* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <linux/kthread.h>
-#include <uapi/linux/lustre_ioctl.h>
+#include <uapi/linux/lustre/lustre_ioctl.h>
#include <lustre_log.h>
#include <lustre_obdo.h>
-#include <lustre_param.h>
+#include <uapi/linux/lustre/lustre_param.h>
#include <obd_class.h>
#include "osp_internal.h"
if (IS_ERR(dto))
RETURN(PTR_ERR(dto));
+ osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &osp->opd_last_id,
+ osp->opd_index);
+
/* object will be released in device cleanup path */
- if (osi->osi_attr.la_size >=
- sizeof(osi->osi_id) * (osp->opd_index + 1)) {
- osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
- osp->opd_index);
+ if (osi->osi_attr.la_size >= (osi->osi_off + osi->osi_lb.lb_len)) {
rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
if (rc != 0 && rc != -EFAULT)
GOTO(out, rc);
+ /* In case of idif bits 32-48 go to f_seq
+ * (see osp_init_last_seq). So don't care
+ * about u64->u32 convertion. */
+ fid->f_oid = osp->opd_last_id;
}
if (rc == -EFAULT) { /* fresh LAST_ID */
+ osp->opd_last_id = 0;
fid->f_oid = 0;
- osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
- osp->opd_index);
rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
osi->osi_off);
if (rc != 0)
if (IS_ERR(dto))
RETURN(PTR_ERR(dto));
+ osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
+ osp->opd_index);
+
/* object will be released in device cleanup path */
- if (osi->osi_attr.la_size >=
- sizeof(osi->osi_id) * (osp->opd_index + 1)) {
- osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
- osp->opd_index);
+ if (osi->osi_attr.la_size >= (osi->osi_off + osi->osi_lb.lb_len)) {
rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
if (rc != 0 && rc != -EFAULT)
GOTO(out, rc);
+ if (fid_is_idif(fid))
+ fid->f_seq = fid_idif_seq(osp->opd_last_id,
+ osp->opd_index);
}
if (rc == -EFAULT) { /* fresh OSP */
fid->f_seq = 0;
- osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
- osp->opd_index);
rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
osi->osi_off);
if (rc != 0)
rc = osp_init_last_seq(env, osp);
if (rc < 0) {
- CERROR("%s: Can not get ids %d from old objid!\n",
+ CERROR("%s: Can not get sequence %d from old objseq!\n",
osp->opd_obd->obd_name, rc);
GOTO(out, rc);
}
GOTO(out, rc = -EINVAL);
}
+ osp_fid_to_obdid(&osp->opd_last_used_fid, &osp->opd_last_id);
CDEBUG(D_INFO, "%s: Init last used fid "DFID"\n",
osp->opd_obd->obd_name, PFID(&osp->opd_last_used_fid));
out:
rc = osp_disconnect(d);
+ osp_statfs_fini(d);
+
if (!d->opd_connect_mdt) {
/* stop sync thread */
osp_sync_fini(d);
static int osp_process_config(const struct lu_env *env,
struct lu_device *dev, struct lustre_cfg *lcfg)
{
- struct osp_device *d = lu2osp_dev(dev);
- struct obd_device *obd = d->opd_obd;
- int rc;
+ struct osp_device *d = lu2osp_dev(dev);
+ struct dt_device *dt = lu2dt_dev(dev);
+ struct obd_device *obd = d->opd_obd;
+ ssize_t count;
+ int rc;
ENTRY;
rc = osp_shutdown(env, d);
break;
case LCFG_PARAM:
- LASSERT(obd);
- rc = class_process_proc_param(d->opd_connect_mdt ?
- PARAM_OSP : PARAM_OSC,
- obd->obd_vars, lcfg, obd);
- if (rc > 0)
- rc = 0;
- if (rc == -ENOSYS) {
- /* class_process_proc_param() haven't found matching
- * parameter and returned ENOSYS so that layer(s)
+ count = class_modify_config(lcfg, d->opd_connect_mdt ?
+ PARAM_OSP : PARAM_OSC,
+ &dt->dd_kobj);
+ if (count < 0) {
+ /* class_modify_config() haven't found matching
+ * parameter and returned an error so that layer(s)
* below could use that. But OSP is the bottom, so
- * just ignore it */
+ * just ignore it
+ */
CERROR("%s: unknown param %s\n",
(char *)lustre_cfg_string(lcfg, 0),
(char *)lustre_cfg_string(lcfg, 1));
- rc = 0;
}
+ rc = 0;
break;
default:
CERROR("%s: unknown command %u\n",
* \retval negative negative errno if get statfs failed.
*/
static int osp_statfs(const struct lu_env *env, struct dt_device *dev,
- struct obd_statfs *sfs)
+ struct obd_statfs *sfs, struct obd_statfs_info *info)
{
struct osp_device *d = dt2osp_dev(dev);
struct obd_import *imp = d->opd_obd->u.cli.cl_import;
if (unlikely(d->opd_imp_active == 0))
RETURN(-ENOTCONN);
+ /* return recently updated data */
+ *sfs = d->opd_statfs;
+ if (info) {
+ info->os_reserved_mb_low = d->opd_reserved_mb_low;
+ info->os_reserved_mb_high = d->opd_reserved_mb_high;
+ }
+
if (d->opd_pre == NULL)
RETURN(0);
- /* return recently updated data */
- *sfs = d->opd_statfs;
+ CDEBUG(D_OTHER, "%s: %llu blocks, %llu free, %llu avail, "
+ "%u reserved mb low, %u reserved mb high,"
+ "%llu files, %llu free files\n", d->opd_obd->obd_name,
+ sfs->os_blocks, sfs->os_bfree, sfs->os_bavail,
+ d->opd_reserved_mb_low, d->opd_reserved_mb_high,
+ sfs->os_files, sfs->os_ffree);
+
+
+ if (info && !info->os_enable_pre)
+ RETURN(0);
/*
- * layer above osp (usually lod) can use ffree to estimate
- * how many objects are available for immediate creation
+ * The layer above osp (usually lod) can use f_precreated to
+ * estimate how many objects are available for immediate usage.
*/
spin_lock(&d->opd_pre_lock);
- LASSERTF(fid_seq(&d->opd_pre_last_created_fid) ==
- fid_seq(&d->opd_pre_used_fid),
- "last_created "DFID", next_fid "DFID"\n",
- PFID(&d->opd_pre_last_created_fid),
- PFID(&d->opd_pre_used_fid));
- sfs->os_fprecreated = fid_oid(&d->opd_pre_last_created_fid) -
- fid_oid(&d->opd_pre_used_fid);
+ sfs->os_fprecreated = osp_fid_diff(&d->opd_pre_last_created_fid,
+ &d->opd_pre_used_fid);
sfs->os_fprecreated -= d->opd_pre_reserved;
LASSERTF(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2,
"last_created "DFID", next_fid "DFID", reserved %llu\n",
PFID(&d->opd_pre_last_created_fid), PFID(&d->opd_pre_used_fid),
d->opd_pre_reserved);
spin_unlock(&d->opd_pre_lock);
-
- CDEBUG(D_OTHER, "%s: %llu blocks, %llu free, %llu avail, "
- "%llu files, %llu free files\n", d->opd_obd->obd_name,
- sfs->os_blocks, sfs->os_bfree, sfs->os_bavail,
- sfs->os_files, sfs->os_ffree);
RETURN(0);
}
static int osp_sync(const struct lu_env *env, struct dt_device *dev)
{
struct osp_device *d = dt2osp_dev(dev);
- cfs_time_t expire;
struct l_wait_info lwi = { 0 };
- unsigned long id, old;
- int rc = 0;
- unsigned long start = cfs_time_current();
+ time64_t start = ktime_get_seconds();
+ int recs, rc = 0;
+ u64 old;
+
ENTRY;
/* No Sync between MDTs yet. */
if (d->opd_connect_mdt)
RETURN(0);
+ recs = atomic_read(&d->opd_sync_changes);
+ old = atomic64_read(&d->opd_sync_processed_recs);
+
+ osp_sync_force(env, dt2osp_dev(dev));
+
if (unlikely(d->opd_imp_active == 0))
RETURN(-ENOTCONN);
- id = d->opd_syn_last_used_id;
down_write(&d->opd_async_updates_rwsem);
CDEBUG(D_OTHER, "%s: async updates %d\n", d->opd_obd->obd_name,
atomic_read(&d->opd_async_updates_count));
/* make sure the connection is fine */
- expire = cfs_time_shift(obd_timeout);
- lwi = LWI_TIMEOUT(expire - cfs_time_current(), osp_sync_timeout, d);
- rc = l_wait_event(d->opd_syn_barrier_waitq,
+ lwi = LWI_TIMEOUT(cfs_time_seconds(obd_timeout), osp_sync_timeout, d);
+ rc = l_wait_event(d->opd_sync_barrier_waitq,
atomic_read(&d->opd_async_updates_count) == 0,
&lwi);
up_write(&d->opd_async_updates_rwsem);
if (rc != 0)
GOTO(out, rc);
- CDEBUG(D_CACHE, "%s: id: used %lu, processed %llu\n",
- d->opd_obd->obd_name, id, d->opd_syn_last_processed_id);
-
- /* wait till all-in-line are processed */
- while (d->opd_syn_last_processed_id < id) {
-
- old = d->opd_syn_last_processed_id;
+ CDEBUG(D_CACHE, "%s: processed %llu\n", d->opd_obd->obd_name,
+ (unsigned long long)atomic64_read(&d->opd_sync_processed_recs));
+ while (atomic64_read(&d->opd_sync_processed_recs) < old + recs) {
+ __u64 last = atomic64_read(&d->opd_sync_processed_recs);
/* make sure the connection is fine */
- expire = cfs_time_shift(obd_timeout);
- lwi = LWI_TIMEOUT(expire - cfs_time_current(),
+ lwi = LWI_TIMEOUT(cfs_time_seconds(obd_timeout),
osp_sync_timeout, d);
- l_wait_event(d->opd_syn_barrier_waitq,
- d->opd_syn_last_processed_id >= id,
- &lwi);
+ l_wait_event(d->opd_sync_barrier_waitq,
+ atomic64_read(&d->opd_sync_processed_recs)
+ >= old + recs, &lwi);
- if (d->opd_syn_last_processed_id >= id)
+ if (atomic64_read(&d->opd_sync_processed_recs) >= old + recs)
break;
- if (d->opd_syn_last_processed_id != old) {
+ if (atomic64_read(&d->opd_sync_processed_recs) != last) {
/* some progress have been made,
* keep trying... */
continue;
}
/* block new processing (barrier>0 - few callers are possible */
- atomic_inc(&d->opd_syn_barrier);
+ atomic_inc(&d->opd_sync_barrier);
CDEBUG(D_CACHE, "%s: %u in flight\n", d->opd_obd->obd_name,
- atomic_read(&d->opd_syn_rpc_in_flight));
+ atomic_read(&d->opd_sync_rpcs_in_flight));
/* wait till all-in-flight are replied, so executed by the target */
/* XXX: this is used by LFSCK at the moment, which doesn't require
* all the changes to be committed, but in general it'd be
* better to wait till commit */
- while (atomic_read(&d->opd_syn_rpc_in_flight) > 0) {
+ while (atomic_read(&d->opd_sync_rpcs_in_flight) > 0) {
+ old = atomic_read(&d->opd_sync_rpcs_in_flight);
- old = atomic_read(&d->opd_syn_rpc_in_flight);
-
- expire = cfs_time_shift(obd_timeout);
- lwi = LWI_TIMEOUT(expire - cfs_time_current(),
+ lwi = LWI_TIMEOUT(cfs_time_seconds(obd_timeout),
osp_sync_timeout, d);
- l_wait_event(d->opd_syn_barrier_waitq,
- atomic_read(&d->opd_syn_rpc_in_flight) == 0,
+ l_wait_event(d->opd_sync_barrier_waitq,
+ atomic_read(&d->opd_sync_rpcs_in_flight) == 0,
&lwi);
- if (atomic_read(&d->opd_syn_rpc_in_flight) == 0)
+ if (atomic_read(&d->opd_sync_rpcs_in_flight) == 0)
break;
- if (atomic_read(&d->opd_syn_rpc_in_flight) != old) {
+ if (atomic_read(&d->opd_sync_rpcs_in_flight) != old) {
/* some progress have been made */
continue;
}
out:
/* resume normal processing (barrier=0) */
- atomic_dec(&d->opd_syn_barrier);
- __osp_sync_check_for_work(d);
+ atomic_dec(&d->opd_sync_barrier);
+ osp_sync_check_for_work(d);
- CDEBUG(D_CACHE, "%s: done in %lu: rc = %d\n", d->opd_obd->obd_name,
- cfs_time_current() - start, rc);
+ CDEBUG(D_CACHE, "%s: done in %lld: rc = %d\n", d->opd_obd->obd_name,
+ ktime_get_seconds() - start, rc);
RETURN(rc);
}
{
struct obd_device *obd;
struct obd_import *imp;
- class_uuid_t uuid;
char *src, *tgt, *mdt, *osdname = NULL;
int rc;
long idx;
strcat(osdname, "-osd");
CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src);
- if (osp->opd_connect_mdt) {
- struct client_obd *cli = &osp->opd_obd->u.cli;
-
- OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock));
- if (!cli->cl_rpc_lock)
- GOTO(out_fini, rc = -ENOMEM);
- osp_init_rpc_lock(cli->cl_rpc_lock);
- }
+ osp_init_rpc_lock(osp);
osp->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
osp->opd_dt_dev.dd_ops = &osp_dt_ops;
GOTO(out_ref, rc);
}
- osp_lprocfs_init(osp);
+ osp_tunables_init(osp);
rc = obd_fid_init(osp->opd_obd, NULL, osp->opd_connect_mdt ?
LUSTRE_SEQ_METADATA : LUSTRE_SEQ_DATA);
if (rc < 0)
GOTO(out_precreat, rc);
} else {
+ osp->opd_got_disconnected = 1;
rc = osp_update_init(osp);
if (rc != 0)
GOTO(out_fid, rc);
}
+ rc = osp_init_statfs(osp);
+ if (rc)
+ GOTO(out_precreat, rc);
+
ns_register_cancel(obd->obd_namespace, osp_cancel_weight);
/*
* Initiate connect to OST
*/
- ll_generate_random_uuid(uuid);
- class_uuid_unparse(uuid, &osp->opd_cluuid);
-
imp = obd->u.cli.cl_import;
rc = ptlrpc_init_import(imp);
out_fid:
obd_fid_fini(osp->opd_obd);
out_proc:
- ptlrpc_lprocfs_unregister_obd(obd);
- lprocfs_obd_cleanup(obd);
- if (osp->opd_symlink)
- lprocfs_remove(&osp->opd_symlink);
+ osp_tunables_fini(osp);
client_obd_cleanup(obd);
out_ref:
ptlrpcd_decref();
out_disconnect:
- if (osp->opd_connect_mdt) {
- struct client_obd *cli = &osp->opd_obd->u.cli;
- if (cli->cl_rpc_lock != NULL) {
- OBD_FREE_PTR(cli->cl_rpc_lock);
- cli->cl_rpc_lock = NULL;
- }
- }
obd_disconnect(osp->opd_storage_exp);
out_fini:
if (osdname)
osp->opd_async_requests = NULL;
}
- if (osp->opd_storage_exp)
+ if (osp->opd_storage_exp) {
+ /* wait for the commit callbacks to complete */
+ wait_event(osp->opd_sync_waitq,
+ atomic_read(&osp->opd_commits_registered) == 0);
obd_disconnect(osp->opd_storage_exp);
-
- if (osp->opd_symlink)
- lprocfs_remove(&osp->opd_symlink);
+ }
LASSERT(osp->opd_obd);
- ptlrpc_lprocfs_unregister_obd(osp->opd_obd);
- lprocfs_obd_cleanup(osp->opd_obd);
-
- if (osp->opd_connect_mdt) {
- struct client_obd *cli = &osp->opd_obd->u.cli;
- if (cli->cl_rpc_lock != NULL) {
- OBD_FREE_PTR(cli->cl_rpc_lock);
- cli->cl_rpc_lock = NULL;
- }
- }
rc = client_obd_cleanup(osp->opd_obd);
if (rc != 0) {
RETURN(ERR_PTR(rc));
}
+ osp_tunables_fini(osp);
+
ptlrpcd_decref();
RETURN(NULL);
* \retval negative negative errno if statfs failed.
*/
static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp,
- struct obd_statfs *osfs, __u64 unused, __u32 flags)
+ struct obd_statfs *osfs, time64_t unused, __u32 flags)
{
struct obd_statfs *msfs;
struct ptlrpc_request *req;
d->opd_imp_connected = 1;
d->opd_imp_seen_connected = 1;
d->opd_obd->obd_inactive = 0;
+ wake_up(&d->opd_pre_waitq);
if (d->opd_connect_mdt)
break;
- if (d->opd_pre != NULL)
- wake_up(&d->opd_pre_waitq);
-
- __osp_sync_check_for_work(d);
+ osp_sync_check_for_work(d);
CDEBUG(D_HA, "got connected\n");
break;
case IMP_EVENT_INVALIDATE:
rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
data->ioc_offset);
break;
- case OBD_IOC_PING_TARGET:
- rc = ptlrpc_obd_ping(obd);
- break;
default:
CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name,
cmd, current_comm());
LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info);
struct lu_context_key osp_txn_key = {
- .lct_tags = LCT_OSP_THREAD | LCT_TX_HANDLE,
+ .lct_tags = LCT_OSP_THREAD,
.lct_init = osp_txn_key_init,
.lct_fini = osp_txn_key_fini
};
.o_fid_alloc = osp_fid_alloc,
};
-struct llog_operations osp_mds_ost_orig_logops;
-
/**
* Initialize OSP module.
*
*/
static int __init osp_init(void)
{
- struct obd_type *type;
+ struct obd_type *sym;
int rc;
rc = lu_kmem_init(osp_caches);
if (rc)
return rc;
-
- rc = class_register_type(&osp_obd_device_ops, NULL, true, NULL,
+ rc = class_register_type(&osp_obd_device_ops, NULL, false, NULL,
LUSTRE_OSP_NAME, &osp_device_type);
if (rc != 0) {
lu_kmem_fini(osp_caches);
return rc;
}
- rc = class_register_type(&lwp_obd_device_ops, NULL, true, NULL,
+ rc = class_register_type(&lwp_obd_device_ops, NULL, false, NULL,
LUSTRE_LWP_NAME, &lwp_device_type);
if (rc != 0) {
class_unregister_type(LUSTRE_OSP_NAME);
return rc;
}
- /* Note: add_rec/delcare_add_rec will be only used by catalogs */
- osp_mds_ost_orig_logops = llog_osd_ops;
- osp_mds_ost_orig_logops.lop_add = llog_cat_add_rec;
- osp_mds_ost_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
-
- /* create "osc" entry in procfs for compatibility purposes */
- type = class_search_type(LUSTRE_OSC_NAME);
- if (type != NULL && type->typ_procroot != NULL)
- return rc;
-
- type = class_search_type(LUSTRE_OSP_NAME);
- type->typ_procsym = lprocfs_register("osc", proc_lustre_root,
- NULL, NULL);
- if (IS_ERR(type->typ_procsym)) {
- CERROR("osp: can't create compat entry \"osc\": %d\n",
- (int) PTR_ERR(type->typ_procsym));
- type->typ_procsym = NULL;
+ /* create "osc" entry for compatibility purposes */
+ sym = class_add_symlinks(LUSTRE_OSC_NAME, true);
+ if (IS_ERR(sym)) {
+ rc = PTR_ERR(sym);
+ /* does real "osc" already exist ? */
+ if (rc == -EEXIST)
+ rc = 0;
}
+
return rc;
}
*/
static void __exit osp_exit(void)
{
+ struct obd_type *sym = class_search_type(LUSTRE_OSC_NAME);
+
+ /* if this was never fully initialized by the osc layer
+ * then we are responsible for freeing this obd_type
+ */
+ if (sym) {
+ /* final put if we manage this obd type */
+ if (sym->typ_sym_filter)
+ kobject_put(&sym->typ_kobj);
+ /* put reference taken by class_search_type */
+ kobject_put(&sym->typ_kobj);
+ }
+
class_unregister_type(LUSTRE_LWP_NAME);
class_unregister_type(LUSTRE_OSP_NAME);
lu_kmem_fini(osp_caches);