#define DEBUG_SUBSYSTEM S_MDS
#include <obd_class.h>
+#include <lustre_fid.h>
#include <lustre_param.h>
+#include <lustre_update.h>
#include "lod_internal.h"
* \param lod LOD to be lookup at.
* \param fid FID of object to find MDT/OST.
* \param tgt MDT/OST index to return.
- * \param flags indidcate the FID is on MDS or OST.
+ * \param type indidcate the FID is on MDS or OST.
**/
int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
- const struct lu_fid *fid, __u32 *tgt, int flags)
+ const struct lu_fid *fid, __u32 *tgt, int type)
{
- struct lu_seq_range range;
+ struct lu_seq_range range = { 0 };
struct lu_server_fld *server_fld;
int rc = 0;
ENTRY;
RETURN(rc);
}
- if (!lod->lod_initialized || !fid_is_norm(fid)) {
+ if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
*tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
RETURN(rc);
}
server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
- range.lsr_flags = flags;
+ fld_range_set_type(&range, type);
rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
if (rc) {
CERROR("%s: Can't find tgt by seq "LPX64", rc %d\n",
}
extern struct lu_object_operations lod_lu_obj_ops;
+extern struct lu_object_operations lod_lu_robj_ops;
extern struct dt_object_operations lod_obj_ops;
/* Slab for OSD object allocation */
const struct lu_object_header *hdr,
struct lu_device *dev)
{
- struct lu_object *lu_obj;
- struct lod_object *lo;
+ struct lod_object *lod_obj;
+ struct lu_object *lu_obj;
+ const struct lu_fid *fid = &hdr->loh_fid;
+ mdsno_t mds;
+ int rc = 0;
+ ENTRY;
- OBD_SLAB_ALLOC_PTR_GFP(lo, lod_object_kmem, CFS_ALLOC_IO);
- if (lo == NULL)
- return NULL;
+ OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, CFS_ALLOC_IO);
+ if (lod_obj == NULL)
+ RETURN(ERR_PTR(-ENOMEM));
- lu_obj = lod2lu_obj(lo);
- dt_object_init(&lo->ldo_obj, NULL, dev);
- lo->ldo_obj.do_ops = &lod_obj_ops;
- lu_obj->lo_ops = &lod_lu_obj_ops;
+ rc = lod_fld_lookup(env, lu2lod_dev(dev), fid, &mds, LU_SEQ_RANGE_MDT);
+ if (rc) {
+ OBD_SLAB_FREE_PTR(lod_obj, lod_object_kmem);
+ RETURN(ERR_PTR(rc));
+ }
- return lu_obj;
+ lod_obj->ldo_mds_num = mds;
+ lu_obj = lod2lu_obj(lod_obj);
+ dt_object_init(&lod_obj->ldo_obj, NULL, dev);
+ lod_obj->ldo_obj.do_ops = &lod_obj_ops;
+ if (likely(mds == lu_site2seq(dev->ld_site)->ss_node_id))
+ lu_obj->lo_ops = &lod_lu_obj_ops;
+ else
+ lu_obj->lo_ops = &lod_lu_robj_ops;
+ RETURN(lu_obj);
}
static int lod_cleanup_desc_tgts(const struct lu_env *env,
return 0;
}
+/*
+ * Init client sequence manager which is used by local MDS to talk to sequence
+ * controller on remote node.
+ */
+static int lod_seq_init_cli(const struct lu_env *env,
+ struct lod_device *lod,
+ char *tgtuuid, int index)
+{
+ struct seq_server_site *ss;
+ struct obd_device *osp;
+ int rc;
+ char *prefix;
+ struct obd_uuid obd_uuid;
+ ENTRY;
+
+ ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
+ LASSERT(ss != NULL);
+
+ /* check if this is adding the first MDC and controller is not yet
+ * initialized. */
+ if (index != 0 || ss->ss_client_seq)
+ RETURN(0);
+
+ obd_str2uuid(&obd_uuid, tgtuuid);
+ osp = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
+ &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
+ if (osp == NULL) {
+ CERROR("%s: can't find %s device\n",
+ lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_name,
+ tgtuuid);
+ RETURN(-EINVAL);
+ }
+
+ if (!osp->obd_set_up) {
+ CERROR("target %s not set up\n", osp->obd_name);
+ rc = -EINVAL;
+ }
+
+ LASSERT(ss->ss_control_exp);
+ OBD_ALLOC_PTR(ss->ss_client_seq);
+ if (ss->ss_client_seq == NULL)
+ RETURN(-ENOMEM);
+
+ OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
+ if (!prefix) {
+ OBD_FREE_PTR(ss->ss_client_seq);
+ ss->ss_client_seq = NULL;
+ RETURN(-ENOMEM);
+ }
+
+ snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", osp->obd_name);
+ rc = seq_client_init(ss->ss_client_seq, ss->ss_control_exp,
+ LUSTRE_SEQ_METADATA, prefix, NULL);
+ OBD_FREE(prefix, MAX_OBD_NAME + 5);
+ if (rc) {
+ OBD_FREE_PTR(ss->ss_client_seq);
+ ss->ss_client_seq = NULL;
+ RETURN(rc);
+ }
+
+ LASSERT(ss->ss_server_seq != NULL);
+ rc = seq_server_set_cli(ss->ss_server_seq, ss->ss_client_seq,
+ env);
+
+ RETURN(rc);
+}
+
+static void lod_seq_fini_cli(struct lod_device *lod)
+{
+ struct seq_server_site *ss;
+
+ ENTRY;
+
+ ss = lu_site2seq(lod2lu_dev(lod)->ld_site);
+ if (ss == NULL) {
+ EXIT;
+ return;
+ }
+
+ if (ss->ss_server_seq)
+ seq_server_set_cli(ss->ss_server_seq,
+ NULL, NULL);
+
+ if (ss->ss_control_exp) {
+ class_export_put(ss->ss_control_exp);
+ ss->ss_control_exp = NULL;
+ }
+
+ EXIT;
+ return;
+}
+
+/**
+ * Procss config log on LOD
+ * \param env environment info
+ * \param dev lod device
+ * \param lcfg config log
+ *
+ * Add osc config log,
+ * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
+ * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
+ * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
+ * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
+ * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
+ * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
+ *
+ * Add mdc config log
+ * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
+ * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
+ * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
+ * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
+ * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
+ * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
+ **/
static int lod_process_config(const struct lu_env *env,
struct lu_device *dev,
struct lustre_cfg *lcfg)
mdt_index = index;
rc = lod_add_device(env, lod, arg1, index, gen,
mdt_index, LUSTRE_MDC_NAME, 1);
+ if (rc == 0)
+ rc = lod_seq_init_cli(env, lod, arg1,
+ mdt_index);
} else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
/*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
mdt_index = 0;
GOTO(out, rc);
}
case LCFG_CLEANUP:
+ case LCFG_PRE_CLEANUP: {
lu_dev_del_linkage(dev->ld_site, dev);
lod_cleanup_desc_tgts(env, lod, &lod->lod_mdt_descs, lcfg);
lod_cleanup_desc_tgts(env, lod, &lod->lod_ost_descs, lcfg);
+
+ lod_seq_fini_cli(lod);
+
+ if (lcfg->lcfg_command == LCFG_PRE_CLEANUP)
+ break;
/*
* do cleanup on underlying storage only when
* all OSPs are cleaned up, as they use that OSD as well
if (rc)
CERROR("error in disconnect from storage: %d\n", rc);
break;
-
+ }
default:
CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
lcfg->lcfg_command);
static struct thandle *lod_trans_create(const struct lu_env *env,
struct dt_device *dev)
{
- return dt_trans_create(env, dt2lod_dev(dev)->lod_child);
+ struct thandle *th;
+
+ th = dt_trans_create(env, dt2lod_dev(dev)->lod_child);
+ if (IS_ERR(th))
+ return th;
+
+ CFS_INIT_LIST_HEAD(&th->th_remote_update_list);
+ return th;
+}
+
+static int lod_remote_sync(const struct lu_env *env, struct dt_device *dev,
+ struct thandle *th)
+{
+ struct update_request *update;
+ int rc = 0;
+ ENTRY;
+
+ if (cfs_list_empty(&th->th_remote_update_list))
+ RETURN(0);
+
+ cfs_list_for_each_entry(update, &th->th_remote_update_list,
+ ur_list) {
+ /* In DNE phase I, there should be only one OSP
+ * here, so we will do send/receive one by one,
+ * instead of sending them parallel, will fix this
+ * in Phase II */
+ th->th_current_request = update;
+ rc = dt_trans_start(env, update->ur_dt, th);
+ if (rc != 0) {
+ /* FIXME how to revert the partial results
+ * once error happened? Resolved by 2 Phase commit */
+ update->ur_rc = rc;
+ break;
+ }
+ }
+
+ RETURN(rc);
}
static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
struct thandle *th)
{
- return dt_trans_start(env, dt2lod_dev(dev)->lod_child, th);
+ struct lod_device *lod = dt2lod_dev((struct dt_device *) dev);
+ int rc;
+
+ rc = lod_remote_sync(env, dev, th);
+ if (rc)
+ return rc;
+
+ return dt_trans_start(env, lod->lod_child, th);
}
static int lod_trans_stop(const struct lu_env *env, struct thandle *th)
{
- /* XXX: we don't know next device, will be fixed with DNE */
- return dt_trans_stop(env, th->th_dev, th);
+ struct update_request *update;
+ struct update_request *tmp;
+ int rc = 0;
+ int rc2 = 0;
+
+ cfs_list_for_each_entry_safe(update, tmp,
+ &th->th_remote_update_list,
+ ur_list) {
+ th->th_current_request = update;
+ rc2 = dt_trans_stop(env, update->ur_dt, th);
+ if (unlikely(rc2 != 0 && rc == 0))
+ rc = rc2;
+ }
+
+ rc2 = dt_trans_stop(env, th->th_dev, th);
+
+ return rc2 != 0 ? rc2 : rc;
}
static void lod_conf_get(const struct lu_env *env,