-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011 Whamcloud, Inc.
- *
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Author: Nathan Rutman <nathan@clusterfs.com>
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_MGC
#define D_MGC D_CONFIG /*|D_WARNING*/
-#ifdef __KERNEL__
-# include <linux/module.h>
-# include <linux/pagemap.h>
-# include <linux/miscdevice.h>
-# include <linux/init.h>
-#else
-# include <liblustre.h>
-#endif
-
+#include <linux/module.h>
#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 <dt_object.h>
+
#include "mgc_internal.h"
static int mgc_name2resid(char *name, int len, struct ldlm_res_id *res_id,
{
__u64 resname = 0;
- if (len > 8) {
+ if (len > sizeof(resname)) {
CERROR("name too long: %s\n", name);
return -EINVAL;
}
case CONFIG_T_SPTLRPC:
resname = 0;
break;
- case CONFIG_T_RECOVER:
+ case CONFIG_T_RECOVER:
+ case CONFIG_T_PARAMS:
resname = type;
break;
default:
int mgc_logname2resid(char *logname, struct ldlm_res_id *res_id, int type)
{
- 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, type);
+ char *name_end;
+ int len;
+
+ /* logname consists of "fsname-nodetype".
+ * e.g. "lustre-MDT0001", "SUN-000-client"
+ * there is an exception: llog "params" */
+ name_end = strrchr(logname, '-');
+ if (!name_end)
+ len = strlen(logname);
+ else
+ len = name_end - logname;
+ return mgc_name2resid(logname, len, res_id, type);
}
/********************** config llog list **********************/
static CFS_LIST_HEAD(config_llog_list);
-static cfs_spinlock_t config_list_lock = CFS_SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(config_list_lock);
/* Take a reference to a config log */
static int config_log_get(struct config_llog_data *cld)
/* spinlock to make sure no item with 0 refcount in the list */
if (cfs_atomic_dec_and_lock(&cld->cld_refcount, &config_list_lock)) {
cfs_list_del(&cld->cld_list_chain);
- cfs_spin_unlock(&config_list_lock);
+ spin_unlock(&config_list_lock);
CDEBUG(D_MGC, "dropping config log %s\n", cld->cld_logname);
config_log_put(cld->cld_recover);
if (cld->cld_sptlrpc)
config_log_put(cld->cld_sptlrpc);
+ if (cld->cld_params)
+ config_log_put(cld->cld_params);
if (cld_is_sptlrpc(cld))
sptlrpc_conf_log_stop(cld->cld_logname);
LASSERT(logname != NULL);
instance = cfg ? cfg->cfg_instance : NULL;
- cfs_spin_lock(&config_list_lock);
+ spin_lock(&config_list_lock);
cfs_list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
/* check if instance equals */
if (instance != cld->cld_cfg.cfg_instance)
cfs_atomic_inc(&found->cld_refcount);
LASSERT(found->cld_stopping == 0 || cld_is_sptlrpc(found) == 0);
}
- cfs_spin_unlock(&config_list_lock);
- RETURN(found);
+ spin_unlock(&config_list_lock);
+ RETURN(found);
}
static
strcpy(cld->cld_logname, logname);
if (cfg)
cld->cld_cfg = *cfg;
- cfs_mutex_init(&cld->cld_lock);
+ else
+ cld->cld_cfg.cfg_callback = class_config_llog_handler;
+ mutex_init(&cld->cld_lock);
cld->cld_cfg.cfg_last_idx = 0;
cld->cld_cfg.cfg_flags = 0;
cld->cld_cfg.cfg_sb = sb;
rc = mgc_logname2resid(logname, &cld->cld_resid, type);
- cfs_spin_lock(&config_list_lock);
- cfs_list_add(&cld->cld_list_chain, &config_llog_list);
- cfs_spin_unlock(&config_list_lock);
+ spin_lock(&config_list_lock);
+ cfs_list_add(&cld->cld_list_chain, &config_llog_list);
+ spin_unlock(&config_list_lock);
if (rc) {
config_log_put(cld);
if (cld_is_sptlrpc(cld)) {
rc = mgc_process_log(obd, cld);
- if (rc)
+ if (rc && rc != -ENOENT)
CERROR("failed processing sptlrpc log: %d\n", rc);
}
struct config_llog_data *cld;
char logname[32];
- if ((lsi->lsi_flags & LSI_SERVER) && IS_OST(lsi->lsi_ldd))
+ if (IS_OST(lsi))
return NULL;
+ /* for osp-on-ost, see lustre_start_osp() */
+ if (IS_MDT(lsi) && lcfg.cfg_instance)
+ return NULL;
+
/* we have to use different llog for clients and mdts for cmd
* where only clients are notified if one of cmd server restarts */
LASSERT(strlen(fsname) < sizeof(logname) / 2);
strcpy(logname, fsname);
- if (lsi->lsi_flags & LSI_SERVER) { /* mdt */
+ if (IS_SERVER(lsi)) { /* mdt */
LASSERT(lcfg.cfg_instance == NULL);
lcfg.cfg_instance = sb;
strcat(logname, "-mdtir");
return cld;
}
+static struct config_llog_data *config_params_log_add(struct obd_device *obd,
+ struct config_llog_instance *cfg, struct super_block *sb)
+{
+ struct config_llog_instance lcfg = *cfg;
+ struct config_llog_data *cld;
+
+ lcfg.cfg_instance = sb;
+
+ cld = do_config_log_add(obd, PARAMS_FILENAME, CONFIG_T_PARAMS,
+ &lcfg, sb);
+
+ return cld;
+}
/** Add this log to the list of active logs watched by an MGC.
* Active means we're watching for updates.
struct config_llog_instance *cfg,
struct super_block *sb)
{
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct config_llog_data *cld;
- struct config_llog_data *sptlrpc_cld;
- char seclogname[32];
- char *ptr;
- ENTRY;
+ struct lustre_sb_info *lsi = s2lsi(sb);
+ struct config_llog_data *cld;
+ struct config_llog_data *sptlrpc_cld;
+ struct config_llog_data *params_cld;
+ char seclogname[32];
+ char *ptr;
+ int rc;
+ ENTRY;
CDEBUG(D_MGC, "adding config log %s:%p\n", logname, cfg->cfg_instance);
CONFIG_T_SPTLRPC, NULL, NULL);
if (IS_ERR(sptlrpc_cld)) {
CERROR("can't create sptlrpc log: %s\n", seclogname);
- RETURN(PTR_ERR(sptlrpc_cld));
+ GOTO(out_err, rc = PTR_ERR(sptlrpc_cld));
}
}
+ params_cld = config_params_log_add(obd, cfg, sb);
+ if (IS_ERR(params_cld)) {
+ rc = PTR_ERR(params_cld);
+ CERROR("%s: can't create params log: rc = %d\n",
+ obd->obd_name, rc);
+ GOTO(out_err1, rc);
+ }
- cld = do_config_log_add(obd, logname, CONFIG_T_CONFIG, cfg, sb);
- if (IS_ERR(cld)) {
- CERROR("can't create log: %s\n", logname);
- config_log_put(sptlrpc_cld);
- RETURN(PTR_ERR(cld));
- }
+ cld = do_config_log_add(obd, logname, CONFIG_T_CONFIG, cfg, sb);
+ if (IS_ERR(cld)) {
+ CERROR("can't create log: %s\n", logname);
+ GOTO(out_err2, rc = PTR_ERR(cld));
+ }
- cld->cld_sptlrpc = sptlrpc_cld;
+ cld->cld_sptlrpc = sptlrpc_cld;
+ cld->cld_params = params_cld;
LASSERT(lsi->lsi_lmd);
if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)) {
struct config_llog_data *recover_cld;
- *strrchr(seclogname, '-') = 0;
+ ptr = strrchr(seclogname, '-');
+ if (ptr != NULL) {
+ *ptr = 0;
+ }
+ else {
+ CERROR("sptlrpc log name not correct: %s", seclogname);
+ config_log_put(cld);
+ RETURN(-EINVAL);
+ }
recover_cld = config_recover_log_add(obd, seclogname, cfg, sb);
- if (IS_ERR(recover_cld)) {
- config_log_put(cld);
- RETURN(PTR_ERR(recover_cld));
- }
- cld->cld_recover = recover_cld;
- }
+ if (IS_ERR(recover_cld))
+ GOTO(out_err3, rc = PTR_ERR(recover_cld));
+ cld->cld_recover = recover_cld;
+ }
- RETURN(0);
+ RETURN(0);
+
+out_err3:
+ config_log_put(cld);
+
+out_err2:
+ config_log_put(params_cld);
+
+out_err1:
+ config_log_put(sptlrpc_cld);
+
+out_err:
+ RETURN(rc);
}
-CFS_DECLARE_MUTEX(llog_process_lock);
+DEFINE_MUTEX(llog_process_lock);
/** Stop watching for updates on this log.
*/
{
struct config_llog_data *cld;
struct config_llog_data *cld_sptlrpc = NULL;
+ struct config_llog_data *cld_params = NULL;
struct config_llog_data *cld_recover = NULL;
int rc = 0;
ENTRY;
if (cld == NULL)
RETURN(-ENOENT);
- cfs_mutex_lock(&cld->cld_lock);
+ 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:
* calling start_log.
*/
if (unlikely(cld->cld_stopping)) {
- cfs_mutex_unlock(&cld->cld_lock);
+ mutex_unlock(&cld->cld_lock);
/* drop the ref from the find */
config_log_put(cld);
RETURN(rc);
cld_recover = cld->cld_recover;
cld->cld_recover = NULL;
- cfs_mutex_unlock(&cld->cld_lock);
-
- if (cld_recover) {
- cfs_mutex_lock(&cld_recover->cld_lock);
- cld_recover->cld_stopping = 1;
- cfs_mutex_unlock(&cld_recover->cld_lock);
- config_log_put(cld_recover);
- }
-
- cfs_spin_lock(&config_list_lock);
- cld_sptlrpc = cld->cld_sptlrpc;
- cld->cld_sptlrpc = NULL;
- cfs_spin_unlock(&config_list_lock);
+ mutex_unlock(&cld->cld_lock);
+
+ if (cld_recover) {
+ mutex_lock(&cld_recover->cld_lock);
+ cld_recover->cld_stopping = 1;
+ mutex_unlock(&cld_recover->cld_lock);
+ config_log_put(cld_recover);
+ }
+
+ spin_lock(&config_list_lock);
+ cld_sptlrpc = cld->cld_sptlrpc;
+ cld->cld_sptlrpc = NULL;
+ cld_params = cld->cld_params;
+ cld->cld_params = NULL;
+ spin_unlock(&config_list_lock);
if (cld_sptlrpc)
config_log_put(cld_sptlrpc);
+ if (cld_params) {
+ mutex_lock(&cld_params->cld_lock);
+ cld_params->cld_stopping = 1;
+ mutex_unlock(&cld_params->cld_lock);
+ config_log_put(cld_params);
+ }
+
/* drop the ref from the find */
config_log_put(cld);
/* drop the start ref */
ENTRY;
rc = snprintf(page, count, "imperative_recovery: %s\n",
- OCD_HAS_FLAG(ocd, IMP_RECOV) ? "ON" : "OFF");
+ OCD_HAS_FLAG(ocd, IMP_RECOV) ? "ENABLED" : "DISABLED");
rc += snprintf(page + rc, count - rc, "client_state:\n");
- cfs_spin_lock(&config_list_lock);
- cfs_list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
- if (cld->cld_recover == NULL)
- continue;
- rc += snprintf(page + rc, count - rc,
- " - { client: %s, nidtbl_version: %u }\n",
- cld->cld_logname,
- cld->cld_recover->cld_cfg.cfg_last_idx);
- }
- cfs_spin_unlock(&config_list_lock);
-
- RETURN(rc);
+ spin_lock(&config_list_lock);
+ cfs_list_for_each_entry(cld, &config_llog_list, cld_list_chain) {
+ if (cld->cld_recover == NULL)
+ continue;
+ rc += snprintf(page + rc, count - rc,
+ " - { client: %s, nidtbl_version: %u }\n",
+ cld->cld_logname,
+ cld->cld_recover->cld_cfg.cfg_last_idx);
+ }
+ spin_unlock(&config_list_lock);
+
+ RETURN(rc);
}
/* reenqueue any lost locks */
#define RQ_LATER 0x4
#define RQ_STOP 0x8
static int rq_state = 0;
-static cfs_waitq_t rq_waitq;
-static CFS_DECLARE_COMPLETION(rq_exit);
+static wait_queue_head_t rq_waitq;
+static DECLARE_COMPLETION(rq_exit);
static void do_requeue(struct config_llog_data *cld)
{
+ int rc = 0;
ENTRY;
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);
+ 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);
+ rc = mgc_process_log(cld->cld_mgcexp->exp_obd, cld);
+ if (rc && rc != -ENOENT)
+ CERROR("failed processing log: %d\n", rc);
+ } else {
+ CDEBUG(D_MGC, "disconnecting, won't update log %s\n",
+ cld->cld_logname);
+ }
+ up_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem);
EXIT;
}
static int mgc_requeue_thread(void *data)
{
- char name[] = "ll_cfg_requeue";
int rc = 0;
ENTRY;
- cfs_daemonize(name);
-
CDEBUG(D_MGC, "Starting requeue thread\n");
/* Keep trying failed locks periodically */
- cfs_spin_lock(&config_list_lock);
- rq_state |= RQ_RUNNING;
- while (1) {
- struct l_wait_info lwi;
- struct config_llog_data *cld, *cld_prev;
- int rand = cfs_rand() & MGC_TIMEOUT_RAND_CENTISEC;
- int stopped = !!(rq_state & RQ_STOP);
- int to;
-
- /* Any new or requeued lostlocks will change the state */
- rq_state &= ~(RQ_NOW | RQ_LATER);
- cfs_spin_unlock(&config_list_lock);
-
- /* 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. */
- to = MGC_TIMEOUT_MIN_SECONDS * CFS_HZ;
- to += rand * CFS_HZ / 100; /* rand is centi-seconds */
- lwi = LWI_TIMEOUT(to, NULL, NULL);
- l_wait_event(rq_waitq, rq_state & RQ_STOP, &lwi);
+ spin_lock(&config_list_lock);
+ rq_state |= RQ_RUNNING;
+ while (1) {
+ struct l_wait_info lwi;
+ struct config_llog_data *cld, *cld_prev;
+ int rand = cfs_rand() & MGC_TIMEOUT_RAND_CENTISEC;
+ int stopped = !!(rq_state & RQ_STOP);
+ int to;
+
+ /* Any new or requeued lostlocks will change the state */
+ rq_state &= ~(RQ_NOW | RQ_LATER);
+ spin_unlock(&config_list_lock);
+
+ /* 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. */
+ to = MGC_TIMEOUT_MIN_SECONDS * HZ;
+ to += rand * HZ / 100; /* rand is centi-seconds */
+ lwi = LWI_TIMEOUT(to, NULL, NULL);
+ l_wait_event(rq_waitq, rq_state & RQ_STOP, &lwi);
/*
* iterate & processing through the list. for each cld, process
*/
cld_prev = NULL;
- cfs_spin_lock(&config_list_lock);
- cfs_list_for_each_entry(cld, &config_llog_list,
- cld_list_chain) {
- if (!cld->cld_lostlock)
- continue;
+ spin_lock(&config_list_lock);
+ cfs_list_for_each_entry(cld, &config_llog_list,
+ cld_list_chain) {
+ if (!cld->cld_lostlock)
+ continue;
- cfs_spin_unlock(&config_list_lock);
+ spin_unlock(&config_list_lock);
LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0);
cld_prev = cld;
cld->cld_lostlock = 0;
- if (likely(!stopped))
- do_requeue(cld);
-
- cfs_spin_lock(&config_list_lock);
- }
- cfs_spin_unlock(&config_list_lock);
- if (cld_prev)
- config_log_put(cld_prev);
-
- /* break after scanning the list so that we can drop
- * refcount to losing lock clds */
- if (unlikely(stopped)) {
- cfs_spin_lock(&config_list_lock);
- break;
- }
-
- /* Wait a bit to see if anyone else needs a requeue */
- lwi = (struct l_wait_info) { 0 };
- l_wait_event(rq_waitq, rq_state & (RQ_NOW | RQ_STOP),
- &lwi);
- cfs_spin_lock(&config_list_lock);
- }
- /* spinlock and while guarantee RQ_NOW and RQ_LATER are not set */
- rq_state &= ~RQ_RUNNING;
- cfs_spin_unlock(&config_list_lock);
-
- cfs_complete(&rq_exit);
-
- CDEBUG(D_MGC, "Ending requeue thread\n");
- RETURN(rc);
+ if (likely(!stopped))
+ do_requeue(cld);
+
+ spin_lock(&config_list_lock);
+ }
+ spin_unlock(&config_list_lock);
+ if (cld_prev)
+ config_log_put(cld_prev);
+
+ /* break after scanning the list so that we can drop
+ * refcount to losing lock clds */
+ if (unlikely(stopped)) {
+ spin_lock(&config_list_lock);
+ break;
+ }
+
+ /* Wait a bit to see if anyone else needs a requeue */
+ lwi = (struct l_wait_info) { 0 };
+ l_wait_event(rq_waitq, rq_state & (RQ_NOW | RQ_STOP),
+ &lwi);
+ 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);
+
+ complete(&rq_exit);
+
+ CDEBUG(D_MGC, "Ending requeue thread\n");
+ RETURN(rc);
}
/* Add a cld to the list to requeue. Start the requeue thread if needed.
cld->cld_stopping, rq_state);
LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0);
- cfs_mutex_lock(&cld->cld_lock);
- if (cld->cld_stopping || cld->cld_lostlock) {
- cfs_mutex_unlock(&cld->cld_lock);
- RETURN_EXIT;
- }
- /* this refcount will be released in mgc_requeue_thread. */
- config_log_get(cld);
- cld->cld_lostlock = 1;
- cfs_mutex_unlock(&cld->cld_lock);
-
- /* Hold lock for rq_state */
- cfs_spin_lock(&config_list_lock);
- if (rq_state & RQ_STOP) {
- cfs_spin_unlock(&config_list_lock);
- cld->cld_lostlock = 0;
- config_log_put(cld);
- } else {
- rq_state |= RQ_NOW;
- cfs_spin_unlock(&config_list_lock);
- cfs_waitq_signal(&rq_waitq);
- }
- EXIT;
+ mutex_lock(&cld->cld_lock);
+ if (cld->cld_stopping || cld->cld_lostlock) {
+ mutex_unlock(&cld->cld_lock);
+ RETURN_EXIT;
+ }
+ /* this refcount will be released in mgc_requeue_thread. */
+ config_log_get(cld);
+ cld->cld_lostlock = 1;
+ mutex_unlock(&cld->cld_lock);
+
+ /* Hold lock for rq_state */
+ spin_lock(&config_list_lock);
+ if (rq_state & RQ_STOP) {
+ spin_unlock(&config_list_lock);
+ cld->cld_lostlock = 0;
+ config_log_put(cld);
+ } else {
+ rq_state |= RQ_NOW;
+ spin_unlock(&config_list_lock);
+ wake_up(&rq_waitq);
+ }
+ EXIT;
}
/********************** class fns **********************/
-
-static int mgc_fs_setup(struct obd_device *obd, struct super_block *sb,
- struct vfsmount *mnt)
+static int mgc_local_llog_init(const struct lu_env *env,
+ struct obd_device *obd,
+ struct obd_device *disk)
{
- struct lvfs_run_ctxt saved;
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct client_obd *cli = &obd->u.cli;
- struct dentry *dentry;
- char *label;
- int err = 0;
- ENTRY;
+ struct llog_ctxt *ctxt;
+ int rc;
- LASSERT(lsi);
- LASSERT(lsi->lsi_srv_mnt == mnt);
+ ENTRY;
- /* The mgc fs exclusion sem. Only one fs can be setup at a time. */
- cfs_down(&cli->cl_mgc_sem);
+ rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, disk,
+ &llog_osd_ops);
+ if (rc)
+ RETURN(rc);
- cfs_cleanup_group_info();
+ ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
+ LASSERT(ctxt);
+ ctxt->loc_dir = obd->u.cli.cl_mgc_configs_dir;
+ llog_ctxt_put(ctxt);
- obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
- if (IS_ERR(obd->obd_fsops)) {
- 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));
- }
+ RETURN(0);
+}
- cli->cl_mgc_vfsmnt = mnt;
- fsfilt_setup(obd, mnt->mnt_sb);
-
- OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
- obd->obd_lvfs_ctxt.pwdmnt = mnt;
- obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
- obd->obd_lvfs_ctxt.fs = get_ds();
-
- push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- dentry = ll_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",
- MOUNT_CONFIGS_DIR, err);
- GOTO(err_ops, err);
- }
- cli->cl_mgc_configs_dir = dentry;
+static int mgc_local_llog_fini(const struct lu_env *env,
+ struct obd_device *obd)
+{
+ struct llog_ctxt *ctxt;
- /* We take an obd ref to insure that we can't get to mgc_cleanup
- without calling mgc_fs_cleanup first. */
- class_incref(obd, "mgc_fs", obd);
+ ENTRY;
- label = fsfilt_get_label(obd, mnt->mnt_sb);
- if (label)
- CDEBUG(D_MGC, "MGC using disk labelled=%s\n", label);
+ ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
+ llog_cleanup(env, ctxt);
- /* We keep the cl_mgc_sem until mgc_fs_cleanup */
- RETURN(0);
+ RETURN(0);
+}
-err_ops:
- fsfilt_put_ops(obd->obd_fsops);
- obd->obd_fsops = NULL;
- cli->cl_mgc_vfsmnt = NULL;
- cfs_up(&cli->cl_mgc_sem);
- RETURN(err);
+static int mgc_fs_setup(struct obd_device *obd, struct super_block *sb)
+{
+ struct lustre_sb_info *lsi = s2lsi(sb);
+ struct client_obd *cli = &obd->u.cli;
+ struct lu_fid rfid, fid;
+ struct dt_object *root, *dto;
+ struct lu_env *env;
+ int rc = 0;
+
+ ENTRY;
+
+ LASSERT(lsi);
+ LASSERT(lsi->lsi_dt_dev);
+
+ OBD_ALLOC_PTR(env);
+ if (env == NULL)
+ RETURN(-ENOMEM);
+
+ /* The mgc fs exclusion sem. Only one fs can be setup at a time. */
+ down(&cli->cl_mgc_sem);
+
+ cfs_cleanup_group_info();
+
+ /* Setup the configs dir */
+ rc = lu_env_init(env, LCT_MG_THREAD);
+ if (rc)
+ GOTO(out_err, rc);
+
+ fid.f_seq = FID_SEQ_LOCAL_NAME;
+ fid.f_oid = 1;
+ fid.f_ver = 0;
+ rc = local_oid_storage_init(env, lsi->lsi_dt_dev, &fid,
+ &cli->cl_mgc_los);
+ if (rc)
+ GOTO(out_env, rc);
+
+ rc = dt_root_get(env, lsi->lsi_dt_dev, &rfid);
+ if (rc)
+ GOTO(out_env, rc);
+
+ root = dt_locate_at(env, lsi->lsi_dt_dev, &rfid,
+ &cli->cl_mgc_los->los_dev->dd_lu_dev);
+ if (unlikely(IS_ERR(root)))
+ GOTO(out_los, rc = PTR_ERR(root));
+
+ dto = local_file_find_or_create(env, cli->cl_mgc_los, root,
+ MOUNT_CONFIGS_DIR,
+ S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO);
+ lu_object_put_nocache(env, &root->do_lu);
+ if (IS_ERR(dto))
+ GOTO(out_los, rc = PTR_ERR(dto));
+
+ cli->cl_mgc_configs_dir = dto;
+
+ LASSERT(lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt);
+ rc = mgc_local_llog_init(env, obd, lsi->lsi_osd_exp->exp_obd);
+ if (rc)
+ GOTO(out_llog, rc);
+
+ /* We take an obd ref to insure that we can't get to mgc_cleanup
+ * without calling mgc_fs_cleanup first. */
+ class_incref(obd, "mgc_fs", obd);
+
+ /* We keep the cl_mgc_sem until mgc_fs_cleanup */
+ EXIT;
+out_llog:
+ if (rc) {
+ lu_object_put(env, &cli->cl_mgc_configs_dir->do_lu);
+ cli->cl_mgc_configs_dir = NULL;
+ }
+out_los:
+ if (rc < 0) {
+ local_oid_storage_fini(env, cli->cl_mgc_los);
+ cli->cl_mgc_los = NULL;
+ up(&cli->cl_mgc_sem);
+ }
+out_env:
+ lu_env_fini(env);
+out_err:
+ OBD_FREE_PTR(env);
+ return rc;
}
static int mgc_fs_cleanup(struct obd_device *obd)
{
- struct client_obd *cli = &obd->u.cli;
- int rc = 0;
- ENTRY;
+ struct lu_env env;
+ struct client_obd *cli = &obd->u.cli;
+ int rc;
- LASSERT(cli->cl_mgc_vfsmnt != NULL);
+ ENTRY;
- if (cli->cl_mgc_configs_dir != NULL) {
- 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;
- pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- class_decref(obd, "mgc_fs", obd);
- }
+ LASSERT(cli->cl_mgc_los != NULL);
- cli->cl_mgc_vfsmnt = NULL;
- if (obd->obd_fsops)
- fsfilt_put_ops(obd->obd_fsops);
+ rc = lu_env_init(&env, LCT_MG_THREAD);
+ if (rc)
+ GOTO(unlock, rc);
- cfs_up(&cli->cl_mgc_sem);
+ mgc_local_llog_fini(&env, obd);
- RETURN(rc);
+ lu_object_put_nocache(&env, &cli->cl_mgc_configs_dir->do_lu);
+ cli->cl_mgc_configs_dir = NULL;
+
+ local_oid_storage_fini(&env, cli->cl_mgc_los);
+ cli->cl_mgc_los = NULL;
+ lu_env_fini(&env);
+
+unlock:
+ class_decref(obd, "mgc_fs", obd);
+ up(&cli->cl_mgc_sem);
+
+ RETURN(0);
+}
+
+static int mgc_llog_init(const struct lu_env *env, struct obd_device *obd)
+{
+ struct llog_ctxt *ctxt;
+ int rc;
+
+ ENTRY;
+
+ /* setup only remote ctxt, the local disk context is switched per each
+ * filesystem during mgc_fs_setup() */
+ rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_REPL_CTXT, obd,
+ &llog_client_ops);
+ if (rc)
+ RETURN(rc);
+
+ ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
+ LASSERT(ctxt);
+
+ llog_initiator_connect(ctxt);
+ llog_ctxt_put(ctxt);
+
+ RETURN(0);
}
+static int mgc_llog_fini(const struct lu_env *env, struct obd_device *obd)
+{
+ struct llog_ctxt *ctxt;
+
+ ENTRY;
+
+ ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
+ if (ctxt)
+ llog_cleanup(env, ctxt);
+
+ RETURN(0);
+}
+
+
static cfs_atomic_t mgc_count = CFS_ATOMIC_INIT(0);
static int mgc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
{
if (cfs_atomic_dec_and_test(&mgc_count)) {
int running;
/* stop requeue thread */
- cfs_spin_lock(&config_list_lock);
- running = rq_state & RQ_RUNNING;
- if (running)
- rq_state |= RQ_STOP;
- cfs_spin_unlock(&config_list_lock);
- if (running) {
- cfs_waitq_signal(&rq_waitq);
- cfs_wait_for_completion(&rq_exit);
+ spin_lock(&config_list_lock);
+ running = rq_state & RQ_RUNNING;
+ if (running)
+ rq_state |= RQ_STOP;
+ spin_unlock(&config_list_lock);
+ if (running) {
+ wake_up(&rq_waitq);
+ wait_for_completion(&rq_exit);
}
}
obd_cleanup_client_import(obd);
- rc = obd_llog_finish(obd, 0);
+ rc = mgc_llog_fini(NULL, obd);
if (rc != 0)
CERROR("failed to cleanup llogging subsystems\n");
break;
static int mgc_cleanup(struct obd_device *obd)
{
- struct client_obd *cli = &obd->u.cli;
int rc;
ENTRY;
- LASSERT(cli->cl_mgc_vfsmnt == NULL);
-
/* COMPAT_146 - old config logs may have added profiles we don't
know about */
if (obd->obd_type->typ_refcnt <= 1)
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, lcfg);
- if (rc)
- GOTO(err_decref, rc);
-
- rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
- if (rc) {
- CERROR("failed to setup llogging subsystems\n");
- GOTO(err_cleanup, rc);
- }
-
- lprocfs_mgc_init_vars(&lvars);
- lprocfs_obd_setup(obd, lvars.obd_vars);
- sptlrpc_lprocfs_cliobd_attach(obd);
-
- if (cfs_atomic_inc_return(&mgc_count) == 1) {
- rq_state = 0;
- cfs_waitq_init(&rq_waitq);
-
- /* start requeue thread */
- rc = cfs_create_thread(mgc_requeue_thread, NULL,
- CFS_DAEMON_FLAGS);
- if (rc < 0) {
- CERROR("%s: Cannot start requeue thread (%d),"
- "no more log updates!\n",
- obd->obd_name, rc);
- GOTO(err_cleanup, rc);
- }
- /* rc is the pid of mgc_requeue_thread. */
- rc = 0;
- }
-
- RETURN(rc);
+ struct lprocfs_static_vars lvars;
+ int rc;
+ ENTRY;
+
+ rc = ptlrpcd_addref();
+ if (rc < 0)
+ RETURN(rc);
+
+ rc = client_obd_setup(obd, lcfg);
+ if (rc)
+ GOTO(err_decref, rc);
+
+ rc = mgc_llog_init(NULL, obd);
+ if (rc) {
+ CERROR("failed to setup llogging subsystems\n");
+ GOTO(err_cleanup, rc);
+ }
+
+ lprocfs_mgc_init_vars(&lvars);
+ lprocfs_obd_setup(obd, lvars.obd_vars);
+ sptlrpc_lprocfs_cliobd_attach(obd);
+
+ if (cfs_atomic_inc_return(&mgc_count) == 1) {
+ rq_state = 0;
+ init_waitqueue_head(&rq_waitq);
+
+ /* start requeue thread */
+ rc = PTR_ERR(kthread_run(mgc_requeue_thread, NULL,
+ "ll_cfg_requeue"));
+ if (IS_ERR_VALUE(rc)) {
+ CERROR("%s: Cannot start requeue thread (%d),"
+ "no more log updates!\n",
+ obd->obd_name, rc);
+ GOTO(err_cleanup, rc);
+ }
+ /* rc is the task_struct pointer of mgc_requeue_thread. */
+ rc = 0;
+ }
+
+ RETURN(rc);
err_cleanup:
- client_obd_cleanup(obd);
+ client_obd_cleanup(obd);
err_decref:
- ptlrpcd_decref();
- RETURN(rc);
+ ptlrpcd_decref();
+ RETURN(rc);
}
/* based on ll_mdc_blocking_ast */
/* mgs wants the lock, give it up... */
LDLM_DEBUG(lock, "MGC blocking CB");
ldlm_lock2handle(lock, &lockh);
- rc = ldlm_cli_cancel(&lockh);
- break;
- case LDLM_CB_CANCELING:
- /* We've given up the lock, prepare ourselves to update. */
- LDLM_DEBUG(lock, "MGC cancel CB");
+ rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
+ break;
+ 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],
- (char *)&lock->l_resource->lr_name.name[0]);
+ CDEBUG(D_MGC, "Lock res "DLDLMRES" (%.8s)\n",
+ PLDLMRES(lock->l_resource),
+ (char *)&lock->l_resource->lr_name.name[0]);
if (!cld) {
CDEBUG(D_INFO, "missing data, won't requeue\n");
/* 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)
+ __u64 *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, NULL };
- struct ptlrpc_request *req;
- int short_limit = cld_is_sptlrpc(cld);
- int rc;
- ENTRY;
+ struct config_llog_data *cld = (struct config_llog_data *)data;
+ struct ldlm_enqueue_info einfo = {
+ .ei_type = type,
+ .ei_mode = mode,
+ .ei_cb_bl = mgc_blocking_ast,
+ .ei_cb_cp = ldlm_completion_ast,
+ };
+ struct ptlrpc_request *req;
+ int short_limit = cld_is_sptlrpc(cld);
+ int rc;
+ ENTRY;
CDEBUG(D_MGC, "Enqueue for %s (res "LPX64")\n", cld->cld_logname,
cld->cld_resid.name[0]);
LDLM_ENQUEUE);
if (req == NULL)
RETURN(-ENOMEM);
+
+ req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, 0);
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))
+ if (lsi && IS_SERVER(lsi))
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);
+ NULL, 0, LVB_T_NONE, 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);
static void mgc_notify_active(struct obd_device *unused)
{
- /* wakeup mgc_requeue_thread to requeue mgc lock */
- cfs_spin_lock(&config_list_lock);
- rq_state |= RQ_NOW;
- cfs_spin_unlock(&config_list_lock);
- cfs_waitq_signal(&rq_waitq);
+ /* wakeup mgc_requeue_thread to requeue mgc lock */
+ spin_lock(&config_list_lock);
+ rq_state |= RQ_NOW;
+ spin_unlock(&config_list_lock);
+ wake_up(&rq_waitq);
- /* TODO: Help the MGS rebuild nidtbl. -jay */
+ /* TODO: Help the MGS rebuild nidtbl. -jay */
}
/* Send target_reg message to MGS */
RETURN(rc);
}
-int mgc_set_info_async(struct obd_export *exp, obd_count keylen,
- void *key, obd_count vallen, void *val,
- struct ptlrpc_request_set *set)
+int mgc_set_info_async(const struct lu_env *env, struct obd_export *exp,
+ obd_count keylen, void *key, obd_count vallen,
+ void *val, struct ptlrpc_request_set *set)
{
int rc = -EINVAL;
ENTRY;
rc = mgc_target_register(exp, mti);
RETURN(rc);
}
- 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(-EINVAL);
- lsi = s2lsi(sb);
- rc = mgc_fs_setup(exp->exp_obd, sb, lsi->lsi_srv_mnt);
- if (rc) {
- CERROR("set_fs got %d\n", rc);
- }
- RETURN(rc);
- }
- if (KEY_IS(KEY_CLEAR_FS)) {
- if (vallen != 0)
- RETURN(-EINVAL);
- rc = mgc_fs_cleanup(exp->exp_obd);
- if (rc) {
- CERROR("clear_fs got %d\n", rc);
- }
- RETURN(rc);
- }
+ if (KEY_IS(KEY_SET_FS)) {
+ struct super_block *sb = (struct super_block *)val;
+
+ if (vallen != sizeof(struct super_block))
+ RETURN(-EINVAL);
+
+ rc = mgc_fs_setup(exp->exp_obd, sb);
+ RETURN(rc);
+ }
+ 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;
RETURN(rc);
}
-static int mgc_get_info(struct obd_export *exp, __u32 keylen, void *key,
- __u32 *vallen, void *val, struct lov_stripe_md *unused)
+static int mgc_get_info(const struct lu_env *env, struct obd_export *exp,
+ __u32 keylen, void *key, __u32 *vallen, void *val,
+ struct lov_stripe_md *unused)
{
int rc = -EINVAL;
switch (event) {
case IMP_EVENT_DISCON:
/* MGC imports should not wait for recovery */
+ if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV))
+ ptlrpc_pinger_ir_down();
break;
case IMP_EVENT_INACTIVE:
break;
ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
break;
}
- 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;
- mgc_notify_active(obd);
- break;
+ case IMP_EVENT_ACTIVE:
+ CDEBUG(D_INFO, "%s: Reactivating import\n", obd->obd_name);
+ /* Clearing obd_no_recov allows us to continue pinging */
+ obd->obd_no_recov = 0;
+ mgc_notify_active(obd);
+ if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV))
+ ptlrpc_pinger_ir_up();
+ break;
case IMP_EVENT_OCD:
break;
case IMP_EVENT_DEACTIVATE:
RETURN(rc);
}
-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;
-
- 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, olg, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
- &llog_client_ops);
- if (rc == 0) {
- ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
- 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)
-{
- struct llog_ctxt *ctxt;
- int rc = 0, rc2 = 0;
- ENTRY;
-
- 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);
-}
-
enum {
- CONFIG_READ_NRPAGES_INIT = 1 << (20 - CFS_PAGE_SHIFT),
+ CONFIG_READ_NRPAGES_INIT = 1 << (20 - PAGE_CACHE_SHIFT),
CONFIG_READ_NRPAGES = 4
};
-static int mgc_apply_recover_logs(struct obd_device *obd,
- struct config_llog_data *cld,
- __u64 max_version,
- void *data, int datalen)
+static int mgc_apply_recover_logs(struct obd_device *mgc,
+ struct config_llog_data *cld,
+ __u64 max_version,
+ void *data, int datalen, bool mne_swab)
{
struct config_llog_instance *cfg = &cld->cld_cfg;
struct lustre_sb_info *lsi = s2lsi(cfg->cfg_sb);
u64 prev_version = 0;
char *inst;
char *buf;
- int bufsz = CFS_PAGE_SIZE;
+ int bufsz;
int pos;
int rc = 0;
int off = 0;
-
- OBD_ALLOC(buf, CFS_PAGE_SIZE);
- if (buf == NULL)
- return -ENOMEM;
+ ENTRY;
LASSERT(cfg->cfg_instance != NULL);
LASSERT(cfg->cfg_sb == cfg->cfg_instance);
- inst = buf;
- if (!(lsi->lsi_flags & LSI_SERVER)) {
- pos = sprintf(inst, "%p", cfg->cfg_instance);
+
+ OBD_ALLOC(inst, PAGE_CACHE_SIZE);
+ if (inst == NULL)
+ RETURN(-ENOMEM);
+
+ if (!IS_SERVER(lsi)) {
+ pos = snprintf(inst, PAGE_CACHE_SIZE, "%p", cfg->cfg_instance);
+ if (pos >= PAGE_CACHE_SIZE) {
+ OBD_FREE(inst, PAGE_CACHE_SIZE);
+ return -E2BIG;
+ }
} else {
- LASSERT(IS_MDT(lsi->lsi_ldd));
- pos = sprintf(inst, "MDT%04x", lsi->lsi_ldd->ldd_svindex);
+ LASSERT(IS_MDT(lsi));
+ rc = server_name2svname(lsi->lsi_svname, inst, NULL,
+ PAGE_CACHE_SIZE);
+ if (rc) {
+ OBD_FREE(inst, PAGE_CACHE_SIZE);
+ RETURN(-EINVAL);
+ }
+ pos = strlen(inst);
}
- buf += pos + 1;
- bufsz -= pos + 1;
+
+ ++pos;
+ buf = inst + pos;
+ bufsz = PAGE_CACHE_SIZE - pos;
while (datalen > 0) {
int entry_len = sizeof(*entry);
if (datalen < entry_len) /* must have entry_len at least */
break;
- lustre_swab_mgs_nidtbl_entry(entry);
- LASSERT(entry->mne_length <= CFS_PAGE_SIZE);
- if (entry->mne_length < entry_len)
- break;
+ /* Keep this swab for normal mixed endian handling. LU-1644 */
+ if (mne_swab)
+ lustre_swab_mgs_nidtbl_entry(entry);
+ if (entry->mne_length > PAGE_CACHE_SIZE) {
+ CERROR("MNE too large (%u)\n", entry->mne_length);
+ break;
+ }
+
+ if (entry->mne_length < entry_len)
+ break;
off += entry->mne_length;
datalen -= entry->mne_length;
strcpy(obdname, cld->cld_logname);
cname = strrchr(obdname, '-');
if (cname == NULL) {
- CERROR("mgc: invalid logname %s\n", obdname);
+ CERROR("mgc %s: invalid logname %s\n",
+ mgc->obd_name, obdname);
break;
}
/* find the obd by obdname */
obd = class_name2obd(obdname);
if (obd == NULL) {
- CDEBUG(D_INFO, "mgc: cannot find obdname %s\n",
- obdname);
-
+ CDEBUG(D_INFO, "mgc %s: cannot find obdname %s\n",
+ mgc->obd_name, obdname);
+ rc = 0;
/* this is a safe race, when the ost is starting up...*/
continue;
}
pos += sprintf(params, "%s.import=%s", cname, "connection=");
uuid = buf + pos;
+ down_read(&obd->u.cli.cl_sem);
+ if (obd->u.cli.cl_import == NULL) {
+ /* client does not connect to the OST yet */
+ up_read(&obd->u.cli.cl_sem);
+ rc = 0;
+ continue;
+ }
+
/* TODO: iterate all nids to find one */
/* find uuid by nid */
rc = client_import_find_conn(obd->u.cli.cl_import,
entry->u.nids[0],
(struct obd_uuid *)uuid);
+ up_read(&obd->u.cli.cl_sem);
if (rc < 0) {
CERROR("mgc: cannot find uuid by nid %s\n",
libcfs_nid2str(entry->u.nids[0]));
/* continue, even one with error */
}
- OBD_FREE(inst, CFS_PAGE_SIZE);
- return rc;
+ OBD_FREE(inst, PAGE_CACHE_SIZE);
+ RETURN(rc);
}
/**
struct mgs_config_body *body;
struct mgs_config_res *res;
struct ptlrpc_bulk_desc *desc;
- cfs_page_t **pages;
+ struct page **pages;
int nrpages;
bool eof = true;
+ bool mne_swab = false;
int i;
int ealen;
int rc;
GOTO(out, rc = -ENOMEM);
for (i = 0; i < nrpages; i++) {
- pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
+ pages[i] = alloc_page(GFP_IOFS);
if (pages[i] == NULL)
GOTO(out, rc = -ENOMEM);
}
again:
LASSERT(cld_is_recover(cld));
- LASSERT(cfs_mutex_is_locked(&cld->cld_lock));
+ LASSERT(mutex_is_locked(&cld->cld_lock));
req = ptlrpc_request_alloc(class_exp2cliimp(cld->cld_mgcexp),
&RQF_MGS_CONFIG_READ);
if (req == NULL)
body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
LASSERT(body != NULL);
LASSERT(sizeof(body->mcb_name) > strlen(cld->cld_logname));
- strncpy(body->mcb_name, cld->cld_logname, sizeof(body->mcb_name));
+ if (strlcpy(body->mcb_name, cld->cld_logname, sizeof(body->mcb_name))
+ >= sizeof(body->mcb_name))
+ GOTO(out, rc = -E2BIG);
body->mcb_offset = cfg->cfg_last_idx + 1;
body->mcb_type = cld->cld_type;
- body->mcb_bits = CFS_PAGE_SHIFT;
+ body->mcb_bits = PAGE_CACHE_SHIFT;
body->mcb_units = nrpages;
- /* allocate bulk transfer descriptor */
- desc = ptlrpc_prep_bulk_imp(req, nrpages, BULK_PUT_SINK,
- MGS_BULK_PORTAL);
- if (desc == NULL)
- GOTO(out, rc = -ENOMEM);
+ /* allocate bulk transfer descriptor */
+ desc = ptlrpc_prep_bulk_imp(req, nrpages, 1, BULK_PUT_SINK,
+ MGS_BULK_PORTAL);
+ if (desc == NULL)
+ GOTO(out, rc = -ENOMEM);
- for (i = 0; i < nrpages; i++)
- ptlrpc_prep_bulk_page(desc, pages[i], 0, CFS_PAGE_SIZE);
+ for (i = 0; i < nrpages; i++)
+ ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (ealen < 0)
GOTO(out, rc = ealen);
- if (ealen > nrpages << CFS_PAGE_SHIFT)
+ if (ealen > nrpages << PAGE_CACHE_SHIFT)
GOTO(out, rc = -EINVAL);
if (ealen == 0) { /* no logs transferred */
GOTO(out, rc);
}
+ mne_swab = !!ptlrpc_rep_need_swab(req);
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
+ /* This import flag means the server did an extra swab of IR MNE
+ * records (fixed in LU-1252), reverse it here if needed. LU-1644 */
+ if (unlikely(req->rq_import->imp_need_mne_swab))
+ mne_swab = !mne_swab;
+#else
+#warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
+#endif
+
for (i = 0; i < nrpages && ealen > 0; i++) {
int rc2;
void *ptr;
- ptr = cfs_kmap(pages[i]);
- rc2 = mgc_apply_recover_logs(obd, cld, res->mcr_offset, ptr,
- min_t(int, ealen, CFS_PAGE_SIZE));
- cfs_kunmap(pages[i]);
- if (rc2 < 0) {
- CWARN("Process recover log %s error %d\n",
- cld->cld_logname, rc2);
- break;
+ ptr = kmap(pages[i]);
+ rc2 = mgc_apply_recover_logs(obd, cld, res->mcr_offset, ptr,
+ min_t(int, ealen, PAGE_CACHE_SIZE),
+ mne_swab);
+ kunmap(pages[i]);
+ if (rc2 < 0) {
+ CWARN("Process recover log %s error %d\n",
+ cld->cld_logname, rc2);
+ break;
}
- ealen -= CFS_PAGE_SIZE;
+ ealen -= PAGE_CACHE_SIZE;
}
out:
if (rc == 0 && !eof)
goto again;
- if (pages) {
- for (i = 0; i < nrpages; i++) {
- if (pages[i] == NULL)
- break;
- cfs_free_page(pages[i]);
- }
- OBD_FREE(pages, sizeof(*pages) * nrpages);
- }
- return rc;
-}
-
-/* identical to mgs_log_is_empty */
-static int mgc_llog_is_empty(struct obd_device *obd, struct llog_ctxt *ctxt,
- char *name)
-{
- struct lvfs_run_ctxt saved;
- struct llog_handle *llh;
- int rc = 0;
-
- push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- rc = llog_create(ctxt, &llh, NULL, name);
- if (rc == 0) {
- llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
- rc = llog_get_size(llh);
- llog_close(llh);
- }
- pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- /* header is record 1 */
- return(rc <= 1);
-}
-
-static int mgc_copy_handler(struct llog_handle *llh, struct llog_rec_hdr *rec,
- void *data)
-{
- struct llog_rec_hdr local_rec = *rec;
- struct llog_handle *local_llh = (struct llog_handle *)data;
- char *cfg_buf = (char*) (rec + 1);
- struct lustre_cfg *lcfg;
- int rc = 0;
- ENTRY;
-
- /* 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);
-
- 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);
+ if (pages) {
+ for (i = 0; i < nrpages; i++) {
+ if (pages[i] == NULL)
+ break;
+ __free_page(pages[i]);
+ }
+ OBD_FREE(pages, sizeof(*pages) * nrpages);
+ }
+ 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)
+static int mgc_llog_local_copy(const struct lu_env *env,
+ 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, temp_log);
- if (rc)
- GOTO(out, rc);
-
- /* set the log header uuid for fun */
- OBD_ALLOC_PTR(uuid);
- obd_str2uuid(uuid, logname);
- rc = llog_init_handle(local_llh, LLOG_F_IS_PLAIN, uuid);
- OBD_FREE_PTR(uuid);
- if (rc)
- GOTO(out_closel, rc);
-
- /* open remote log */
- rc = llog_create(rctxt, &remote_llh, NULL, logname);
- if (rc)
- GOTO(out_closel, rc);
- rc = llog_init_handle(remote_llh, LLOG_F_IS_PLAIN, NULL);
- if (rc)
- GOTO(out_closer, rc);
-
- /* Copy remote log */
- rc = llog_process(remote_llh, mgc_copy_handler,(void *)local_llh, NULL);
-
-out_closer:
- rc2 = llog_close(remote_llh);
- if (!rc)
- rc = rc2;
-out_closel:
- rc2 = llog_close(local_llh);
- 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);
+ char *temp_log;
+ int rc;
+
+ ENTRY;
+
+ /*
+ * - copy it to backup using llog_backup()
+ * - copy remote llog to logname using llog_backup()
+ * - if failed then move bakup to logname again
+ */
+
+ OBD_ALLOC(temp_log, strlen(logname) + 1);
+ if (!temp_log)
+ RETURN(-ENOMEM);
+ sprintf(temp_log, "%sT", logname);
+
+ /* make a copy of local llog at first */
+ rc = llog_backup(env, obd, lctxt, lctxt, logname, temp_log);
+ if (rc < 0 && rc != -ENOENT)
+ GOTO(out, rc);
+ /* copy remote llog to the local copy */
+ rc = llog_backup(env, obd, rctxt, lctxt, logname, logname);
+ if (rc == -ENOENT) {
+ /* no remote llog, delete local one too */
+ llog_erase(env, lctxt, NULL, logname);
+ } else if (rc < 0) {
+ /* error during backup, get local one back from the copy */
+ llog_backup(env, obd, lctxt, lctxt, temp_log, logname);
out:
- if (rc)
- CERROR("Failed to copy remote log %s (%d)\n", logname, rc);
- OBD_FREE(temp_log, strlen(logname) + 1);
- RETURN(rc);
+ CERROR("%s: failed to copy remote log %s: rc = %d\n",
+ obd->obd_name, logname, rc);
+ }
+ llog_erase(env, lctxt, NULL, temp_log);
+ OBD_FREE(temp_log, strlen(logname) + 1);
+ return rc;
}
/* local_only means it cannot get remote llogs */
static int mgc_process_cfg_log(struct obd_device *mgc,
- struct config_llog_data *cld,
- int local_only)
+ struct config_llog_data *cld, int local_only)
{
- struct llog_ctxt *ctxt, *lctxt = NULL;
- struct client_obd *cli = &mgc->u.cli;
- struct lvfs_run_ctxt *saved_ctxt;
- struct lustre_sb_info *lsi = NULL;
- int rc = 0, must_pop = 0;
- bool sptlrpc_started = false;
-
- ENTRY;
-
- LASSERT(cld);
- LASSERT(cfs_mutex_is_locked(&cld->cld_lock));
-
- /*
- * local copy of sptlrpc log is controlled elsewhere, don't try to
- * read it up here.
- */
- if (cld_is_sptlrpc(cld) && local_only)
- RETURN(0);
-
- if (cld->cld_cfg.cfg_sb)
- lsi = s2lsi(cld->cld_cfg.cfg_sb);
-
- ctxt = llog_get_context(mgc, LLOG_CONFIG_REPL_CTXT);
- if (!ctxt) {
- CERROR("missing llog context\n");
- RETURN(-EINVAL);
- }
-
- OBD_ALLOC_PTR(saved_ctxt);
- if (saved_ctxt == NULL)
- RETURN(-ENOMEM);
-
- lctxt = llog_get_context(mgc, LLOG_CONFIG_ORIG_CTXT);
-
- /* 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) &&
- (lsi->lsi_srv_mnt == cli->cl_mgc_vfsmnt) &&
- !IS_MGS(lsi->lsi_ldd)) {
- push_ctxt(saved_ctxt, &mgc->obd_lvfs_ctxt, NULL);
- must_pop++;
- if (!local_only)
- /* Only try to copy log if we have the lock. */
- rc = mgc_copy_llog(mgc, ctxt, lctxt, cld->cld_logname);
- if (local_only || rc) {
- if (mgc_llog_is_empty(mgc, lctxt, 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);
- }
- 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
- log has. */
- llog_ctxt_put(ctxt);
- ctxt = lctxt;
- lctxt = NULL;
- } else if (local_only) { /* no local log at client side */
- GOTO(out_pop, rc = -EIO);
- }
-
- if (cld_is_sptlrpc(cld)) {
- sptlrpc_conf_log_update_begin(cld->cld_logname);
- sptlrpc_started = true;
- }
-
- /* 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);
- EXIT;
+ struct llog_ctxt *ctxt, *lctxt = NULL;
+ struct client_obd *cli = &mgc->u.cli;
+ struct lustre_sb_info *lsi = NULL;
+ int rc = 0;
+ bool sptlrpc_started = false;
+ struct lu_env *env;
+
+ ENTRY;
+
+ LASSERT(cld);
+ LASSERT(mutex_is_locked(&cld->cld_lock));
+
+ /*
+ * local copy of sptlrpc log is controlled elsewhere, don't try to
+ * read it up here.
+ */
+ if (cld_is_sptlrpc(cld) && local_only)
+ RETURN(0);
+
+ if (cld->cld_cfg.cfg_sb)
+ lsi = s2lsi(cld->cld_cfg.cfg_sb);
+
+ OBD_ALLOC_PTR(env);
+ if (env == NULL)
+ RETURN(-ENOMEM);
+
+ rc = lu_env_init(env, LCT_MG_THREAD);
+ if (rc)
+ GOTO(out_free, rc);
+
+ ctxt = llog_get_context(mgc, LLOG_CONFIG_REPL_CTXT);
+ LASSERT(ctxt);
+
+ lctxt = llog_get_context(mgc, LLOG_CONFIG_ORIG_CTXT);
+
+ /* 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 && IS_SERVER(lsi) && !IS_MGS(lsi) &&
+ cli->cl_mgc_configs_dir != NULL &&
+ lu2dt_dev(cli->cl_mgc_configs_dir->do_lu.lo_dev) ==
+ lsi->lsi_dt_dev) {
+ if (!local_only)
+ /* Only try to copy log if we have the lock. */
+ rc = mgc_llog_local_copy(env, mgc, ctxt, lctxt,
+ cld->cld_logname);
+ if (local_only || rc) {
+ if (llog_is_empty(env, lctxt, 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 = -ENOENT);
+ }
+ 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
+ * log has. */
+ llog_ctxt_put(ctxt);
+ ctxt = lctxt;
+ lctxt = NULL;
+ } else {
+ if (local_only) /* no local log at client side */
+ GOTO(out_pop, rc = -EIO);
+ }
+
+ if (cld_is_sptlrpc(cld)) {
+ sptlrpc_conf_log_update_begin(cld->cld_logname);
+ sptlrpc_started = true;
+ }
+
+ /* 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(env, ctxt, cld->cld_logname,
+ &cld->cld_cfg);
+ EXIT;
out_pop:
- llog_ctxt_put(ctxt);
- if (lctxt)
- llog_ctxt_put(lctxt);
- if (must_pop)
- pop_ctxt(saved_ctxt, &mgc->obd_lvfs_ctxt, NULL);
-
- OBD_FREE_PTR(saved_ctxt);
- /*
- * 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 (sptlrpc_started) {
- LASSERT(cld_is_sptlrpc(cld));
- sptlrpc_conf_log_update_end(cld->cld_logname);
- class_notify_sptlrpc_conf(cld->cld_logname,
- strlen(cld->cld_logname) -
- strlen("-sptlrpc"));
- }
-
- RETURN(rc);
+ __llog_ctxt_put(env, ctxt);
+ if (lctxt)
+ __llog_ctxt_put(env, lctxt);
+
+ /*
+ * 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 (sptlrpc_started) {
+ LASSERT(cld_is_sptlrpc(cld));
+ sptlrpc_conf_log_update_end(cld->cld_logname);
+ class_notify_sptlrpc_conf(cld->cld_logname,
+ strlen(cld->cld_logname) -
+ strlen("-sptlrpc"));
+ }
+
+ lu_env_fini(env);
+out_free:
+ OBD_FREE_PTR(env);
+ return rc;
}
/** Get a config log from the MGS and process it.
int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld)
{
struct lustre_handle lockh = { 0 };
- int rc = 0, rcl, flags = 0;
+ __u64 flags = LDLM_FL_NO_LRU;
+ int rc = 0, rcl;
ENTRY;
LASSERT(cld);
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);
+ mutex_lock(&cld->cld_lock);
+ if (cld->cld_stopping) {
+ mutex_unlock(&cld->cld_lock);
RETURN(0);
}
CDEBUG(D_MGC, "%s: configuration from log '%s' %sed (%d).\n",
mgc->obd_name, cld->cld_logname, rc ? "fail" : "succeed", rc);
- cfs_mutex_unlock(&cld->cld_lock);
+ mutex_unlock(&cld->cld_lock);
/* Now drop the lock so MGS can revoke it */
if (!rcl) {
cld->cld_cfg.cfg_flags |= CFG_F_COMPAT146;
rc = mgc_process_log(obd, cld);
- if (rc == 0 && cld->cld_recover) {
- rc = mgc_process_log(obd, cld->cld_recover);
+ if (rc == 0 && cld->cld_recover != NULL) {
+ if (OCD_HAS_FLAG(&obd->u.cli.cl_import->
+ imp_connect_data, IMP_RECOV)) {
+ rc = mgc_process_log(obd, cld->cld_recover);
+ } else {
+ struct config_llog_data *cir = cld->cld_recover;
+ cld->cld_recover = NULL;
+ config_log_put(cir);
+ }
if (rc)
CERROR("Cannot process recover llog %d\n", rc);
}
+
+ if (rc == 0 && cld->cld_params != NULL) {
+ rc = mgc_process_log(obd, cld->cld_params);
+ if (rc == -ENOENT) {
+ CDEBUG(D_MGC, "There is no params"
+ "config file yet\n");
+ rc = 0;
+ }
+ /* params log is optional */
+ if (rc)
+ CERROR("%s: can't process params llog: rc = %d\n",
+ obd->obd_name, rc);
+ }
config_log_put(cld);
break;
.o_set_info_async = mgc_set_info_async,
.o_get_info = mgc_get_info,
.o_import_event = mgc_import_event,
- .o_llog_init = mgc_llog_init,
- .o_llog_finish = mgc_llog_finish,
.o_process_config = mgc_process_config,
};