Whamcloud - gitweb
b=6427
[fs/lustre-release.git] / lustre / lmv / lmv_obd.c
index f5262a9..d69c956 100644 (file)
@@ -31,6 +31,7 @@
 #include <linux/pagemap.h>
 #include <asm/div64.h>
 #include <linux/seq_file.h>
+#include <linux/namei.h>
 #else
 #include <liblustre.h>
 #endif
 #include <linux/lprocfs_status.h>
 #include <linux/lustre_fsfilt.h>
 #include <linux/obd_lmv.h>
+#include <linux/lustre_lite.h>
 #include "lmv_internal.h"
 
+/* not defined for liblustre building */
+#if !defined(ATOMIC_INIT)
+#define ATOMIC_INIT(val) { (val) }
+#endif
+
 /* object cache. */
 kmem_cache_t *obj_cache;
 atomic_t obj_cache_count = ATOMIC_INIT(0);
@@ -73,8 +80,8 @@ static void lmv_activate_target(struct lmv_obd *lmv,
 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
                               int activate)
 {
-        struct obd_device *obd;
         struct lmv_tgt_desc *tgt;
+        struct obd_device *obd;
         int i, rc = 0;
         ENTRY;
 
@@ -89,7 +96,7 @@ static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
                 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 
-                if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
+                if (obd_uuid_equals(uuid, &tgt->uuid))
                         break;
         }
 
@@ -126,14 +133,15 @@ static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
                       int active, void *data)
 {
-        int rc;
         struct obd_uuid *uuid;
+        int rc;
+        ENTRY;
 
         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
                 CERROR("unexpected notification of %s %s!\n",
                        watched->obd_type->typ_name,
                        watched->obd_name);
-                return -EINVAL;
+                RETURN(-EINVAL);
         }
         uuid = &watched->u.cli.cl_import->imp_target_uuid;
 
@@ -185,7 +193,8 @@ int lmv_detach(struct obd_device *dev)
 /* this is fake connect function. Its purpose is to initialize lmv and say
  * caller that everything is okay. Real connection will be performed later. */
 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
-                       struct obd_uuid *cluuid, unsigned long flags)
+                       struct obd_uuid *cluuid, struct obd_connect_data *data,
+                       unsigned long flags)
 {
 #ifdef __KERNEL__
         struct proc_dir_entry *lmv_proc_dir;
@@ -216,6 +225,8 @@ static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
         lmv->cluuid = *cluuid;
         lmv->connect_flags = flags;
         sema_init(&lmv->init_sem, 1);
+        if (data)
+                memcpy(&lmv->conn_data, data, sizeof(*data));
 
 #ifdef __KERNEL__
         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
@@ -281,14 +292,15 @@ int lmv_check_connect(struct obd_device *obd)
         struct obd_uuid *cluuid;
         struct obd_export *exp;
         int rc, rc2, i;
+        ENTRY;
 
         if (lmv->connected)
-                return 0;
+                RETURN(0);
         
         down(&lmv->init_sem);
         if (lmv->connected) {
                 up(&lmv->init_sem);
-                return 0;
+                RETURN(0);
         }
 
         cluuid = &lmv->cluuid;
@@ -298,16 +310,16 @@ int lmv_check_connect(struct obd_device *obd)
                cluuid->uuid, obd->obd_name);
 
         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
-                struct obd_device *tgt_obd;
-                struct lustre_handle conn = {0, };
                 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
+                struct lustre_handle conn = {0, };
+                struct obd_device *tgt_obd;
 
                 LASSERT(tgts != NULL);
 
                 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME, 
                                                 &obd->obd_uuid);
                 if (!tgt_obd) {
-                        CERROR("Target %s not attached\n", tgts->uuid.uuid);
+                        CERROR("target %s not attached\n", tgts->uuid.uuid);
                         GOTO(out_disc, rc = -EINVAL);
                 }
 
@@ -325,14 +337,14 @@ int lmv_check_connect(struct obd_device *obd)
                         cluuid->uuid);
 
                 if (!tgt_obd->obd_set_up) {
-                        CERROR("Target %s not set up\n", tgts->uuid.uuid);
+                        CERROR("target %s not set up\n", tgts->uuid.uuid);
                         GOTO(out_disc, rc = -EINVAL);
                 }
                 
-                rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid,
+                rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid, &lmv->conn_data,
                                  lmv->connect_flags);
                 if (rc) {
-                        CERROR("Target %s connect error %d\n",
+                        CERROR("target %s connect error %d\n",
                                 tgts->uuid.uuid, rc);
                         GOTO(out_disc, rc);
                 }
@@ -343,7 +355,7 @@ int lmv_check_connect(struct obd_device *obd)
 
                 rc = obd_register_observer(tgt_obd, obd);
                 if (rc) {
-                        CERROR("Target %s register_observer error %d\n",
+                        CERROR("target %s register_observer error %d\n",
                                tgts->uuid.uuid, rc);
                         obd_disconnect(tgts->ltd_exp, 0);
                         GOTO(out_disc, rc);
@@ -388,7 +400,7 @@ int lmv_check_connect(struct obd_device *obd)
         class_export_put(exp);
         lmv->connected = 1;
         up(&lmv->init_sem);
-        return 0;
+        RETURN(0);
 
  out_disc:
         while (i-- > 0) {
@@ -405,10 +417,10 @@ int lmv_check_connect(struct obd_device *obd)
         }
         class_disconnect(exp, 0);
         up(&lmv->init_sem);
-        RETURN (rc);
+        return rc;
 }
 
-static int lmv_disconnect(struct obd_export *exp, int flags)
+static int lmv_disconnect(struct obd_export *exp, unsigned long flags)
 {
         struct obd_device *obd = class_exp2obd(exp);
         struct lmv_obd *lmv = &obd->u.lmv;
@@ -439,6 +451,9 @@ static int lmv_disconnect(struct obd_export *exp, int flags)
 
                 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
 
+                if (mdc_obd)
+                        mdc_obd->obd_no_recov = obd->obd_no_recov;
+
 #ifdef __KERNEL__
                 if (lmv_proc_dir) {
                         struct proc_dir_entry *mdc_symlink;
@@ -453,10 +468,6 @@ static int lmv_disconnect(struct obd_export *exp, int flags)
                         }
                 }
 #endif
-                if (obd->obd_no_recov) {
-                        if (mdc_obd)
-                                mdc_obd->obd_no_recov = 1;
-                }
                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
                         lmv->tgts[i].ltd_exp->exp_obd->obd_name,
                         lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
@@ -541,27 +552,27 @@ static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
         struct lmv_obd *lmv = &obd->u.lmv;
         ENTRY;
 
-        if (lcfg->lcfg_inllen1 < 1) {
+        if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
                 CERROR("LMV setup requires a descriptor\n");
                 RETURN(-EINVAL);
         }
 
-        if (lcfg->lcfg_inllen2 < 1) {
-                CERROR("LMV setup requires an OST UUID list\n");
+        if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) {
+                CERROR("LMV setup requires an MDT UUID list\n");
                 RETURN(-EINVAL);
         }
 
-        desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
-        if (sizeof(*desc) > lcfg->lcfg_inllen1) {
+        desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
+        if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
                 CERROR("descriptor size wrong: %d > %d\n",
-                       (int)sizeof(*desc), lcfg->lcfg_inllen1);
+                       (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
                 RETURN(-EINVAL);
         }
 
-        uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
-        if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
+        uuids = (struct obd_uuid *)lustre_cfg_buf(lcfg, 2);
+        if (sizeof(*uuids) * desc->ld_tgt_count != LUSTRE_CFG_BUFLEN(lcfg, 2)) {
                 CERROR("UUID array size wrong: %u * %u != %u\n",
-                       sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
+                       sizeof(*uuids), desc->ld_tgt_count, LUSTRE_CFG_BUFLEN(lcfg, 2));
                 RETURN(-EINVAL);
         }
 
@@ -590,7 +601,7 @@ static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
                 OBD_FREE(lmv->tgts, lmv->tgts_size);
         }
 
-        tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME, 
+        tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME,
                                         &obd->obd_uuid);
         if (!tgt_obd) {
                 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
@@ -615,36 +626,43 @@ static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
                       unsigned long max_age)
 {
         struct lmv_obd *lmv = &obd->u.lmv;
-        struct obd_statfs temp;
+        struct obd_statfs *temp;
         int rc = 0, i;
         ENTRY;
         
         rc = lmv_check_connect(obd);
         if (rc)
                 RETURN(rc);
+
+        OBD_ALLOC(temp, sizeof(*temp));
+        if (temp == NULL)
+                RETURN(-ENOMEM);
                 
         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
-                if (lmv->tgts[i].ltd_exp == NULL) {
-                        CWARN("%s: NULL export for %d\n", obd->obd_name, i);
+                if (lmv->tgts[i].ltd_exp == NULL)
                         continue;
-                }
 
-                rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, &temp, max_age);
+                rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
                 if (rc) {
-                        CERROR("can't stat MDS #%d (%s)\n", i,
-                               lmv->tgts[i].ltd_exp->exp_obd->obd_name);
-                        RETURN(rc);
+                        CERROR("can't stat MDS #%d (%s), error %d\n", i,
+                               lmv->tgts[i].ltd_exp->exp_obd->obd_name,
+                               rc);
+                        GOTO(out_free_temp, rc);
                 }
                 if (i == 0) {
-                        memcpy(osfs, &temp, sizeof(temp));
+                        memcpy(osfs, temp, sizeof(*temp));
                 } else {
-                        osfs->os_bavail += temp.os_bavail;
-                        osfs->os_blocks += temp.os_blocks;
-                        osfs->os_ffree += temp.os_ffree;
-                        osfs->os_files += temp.os_files;
+                        osfs->os_bavail += temp->os_bavail;
+                        osfs->os_blocks += temp->os_blocks;
+                        osfs->os_ffree += temp->os_ffree;
+                        osfs->os_files += temp->os_files;
                 }
         }
-        RETURN(rc);
+
+        EXIT;
+out_free_temp:
+        OBD_FREE(temp, sizeof(*temp));
+        return rc;
 }
 
 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
@@ -665,8 +683,8 @@ static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
 }
 
 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
-                       unsigned long valid, unsigned int ea_size,
-                       struct ptlrpc_request **request)
+                       __u64 valid, const char *ea_name, int ea_namelen,
+                       unsigned int ea_size, struct ptlrpc_request **request)
 {
         struct obd_device *obd = exp->exp_obd;
         struct lmv_obd *lmv = &obd->u.lmv;
@@ -680,8 +698,9 @@ static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
 
         LASSERT(i < lmv->desc.ld_tgt_count);
 
+
         rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
-                        ea_size, request);
+                        ea_name, ea_namelen, ea_size, request);
         if (rc)
                 RETURN(rc);
         
@@ -690,11 +709,13 @@ static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
         CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
                OLID4(id), obj ? "(splitted)" : "");
 
-        /* if object is splitted, then we loop over all the slaves and gather
+        /*
+         * if object is splitted, then we loop over all the slaves and gather
          * size attribute. In ideal world we would have to gather also mds field
          * from all slaves, as object is spread over the cluster and this is
          * definitely interesting information and it is not good to loss it,
-         * but...*/
+         * but...
+         */
         if (obj) {
                 struct mds_body *body;
 
@@ -837,17 +858,18 @@ int lmv_get_mea_and_update_object(struct obd_export *exp,
         struct ptlrpc_request *req = NULL;
         struct lmv_obj *obj;
         struct lustre_md md;
-        unsigned long valid;
         int mealen, rc;
+        __u64 valid;
+        ENTRY;
 
         md.mea = NULL;
         mealen = MEA_SIZE_LMV(lmv);
         
-        valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
+        valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
 
         /* time to update mea of parent id */
         rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
-                        id, valid, mealen, &req);
+                        id, valid, NULL, 0, mealen, &req);
         if (rc) {
                 CERROR("md_getattr() failed, error %d\n", rc);
                 GOTO(cleanup, rc);
@@ -869,10 +891,11 @@ int lmv_get_mea_and_update_object(struct obd_export *exp,
         lmv_put_obj(obj);
         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
 
+        EXIT;
 cleanup:
         if (req)
                 ptlrpc_req_finished(req);
-        RETURN(rc);
+        return rc;
 }
 
 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
@@ -906,22 +929,23 @@ repeat:
                op_data->name, OLID4(&op_data->id1));
         
         rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp, 
-                       op_data, data, datalen, mode, uid, gid, rdev, request);
+                       op_data, data, datalen, mode, uid, gid, rdev,
+                       request);
         if (rc == 0) {
                 if (*request == NULL)
                         RETURN(rc);
 
                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
                                       sizeof(*body));
-                LASSERT(body != NULL);
+                if (body == NULL)
+                        RETURN(-ENOMEM);
                 
                 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
-                
-/*                LASSERT(body->valid & OBD_MD_MDS ||
-                        body->mds == id_group(&op_data->id1));*/
         } else if (rc == -ERESTART) {
-                /* directory got splitted. time to update local object and
-                 * repeat the request with proper MDS */
+                /*
+                 * directory got splitted. time to update local object and
+                 * repeat the request with proper MDS.
+                 */
                 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
                 if (rc == 0) {
                         ptlrpc_req_finished(*request);
@@ -937,6 +961,7 @@ int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
         struct lmv_obd *lmv = &obd->u.lmv;
         int rc;
         ENTRY;
+        
         rc = lmv_check_connect(obd);
         if (rc)
                 RETURN(rc);
@@ -956,45 +981,99 @@ int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
         struct obd_device *obd = exp->exp_obd;
         struct lmv_obd *lmv = &obd->u.lmv;
         struct mea *mea = data->mea1;
-        struct mdc_op_data data2;
+        struct mdc_op_data *data2;
         int i, rc, mds;
         ENTRY;
 
+        OBD_ALLOC(data2, sizeof(*data2));
+        if (data2 == NULL)
+                RETURN(-ENOMEM);
+        
         LASSERT(mea != NULL);
         for (i = 0; i < mea->mea_count; i++) {
-                memset(&data2, 0, sizeof(data2));
-                data2.id1 = mea->mea_ids[i];
-                mds = id_group(&data2.id1);
+                memset(data2, 0, sizeof(*data2));
+                data2->id1 = mea->mea_ids[i];
+                mds = id_group(&data2->id1);
                 
                 if (lmv->tgts[mds].ltd_exp == NULL)
                         continue;
 
                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it, 
-                                lockmode, &data2, lockh + i, lmm, lmmsize, 
+                                lockmode, data2, lockh + i, lmm, lmmsize, 
                                 cb_compl, cb_blocking, cb_data);
                 
                 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
-                       OLID4(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
+                       OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status);
                 if (rc)
                         GOTO(cleanup, rc);
-                if (it->d.lustre.it_data) {
+                if (LUSTRE_IT(it)->it_data) {
                         struct ptlrpc_request *req;
-                        req = (struct ptlrpc_request *) it->d.lustre.it_data;
+                        req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data;
                         ptlrpc_req_finished(req);
                 }
                 
-                if (it->d.lustre.it_status)
-                        GOTO(cleanup, rc = it->d.lustre.it_status);
+                if (LUSTRE_IT(it)->it_status)
+                        GOTO(cleanup, rc = LUSTRE_IT(it)->it_status);
         }
-        RETURN(0);
         
+        OBD_FREE(data2, sizeof(*data2));
+        RETURN(0);
 cleanup:
+        OBD_FREE(data2, sizeof(*data2));
+        
         /* drop all taken locks */
         while (--i >= 0) {
                 if (lockh[i].cookie)
                         ldlm_lock_decref(lockh + i, lockmode);
                 lockh[i].cookie = 0;
         }
+        return rc;
+}
+
+int lmv_enqueue_remote(struct obd_export *exp, int lock_type,
+                       struct lookup_intent *it, int lock_mode,
+                       struct mdc_op_data *data, struct lustre_handle *lockh,
+                       void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
+                       ldlm_blocking_callback cb_blocking, void *cb_data)
+{
+        struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
+        struct obd_device *obd = exp->exp_obd;
+        struct lmv_obd *lmv = &obd->u.lmv;
+        struct lustre_handle plock;
+        struct mdc_op_data rdata;
+        struct mds_body *body = NULL;
+        int rc = 0, pmode;
+        ENTRY;
+
+        body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
+        LASSERT(body != NULL);
+
+        if (!(body->valid & OBD_MD_MDS))
+                RETURN(0);
+
+        CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4" -> "DLID4"\n",
+               LL_IT2STR(it), OLID4(&data->id1), OLID4(&body->id1));
+
+        /* we got LOOKUP lock, but we really need attrs */
+        pmode = LUSTRE_IT(it)->it_lock_mode;
+        LASSERT(pmode != 0);
+        memcpy(&plock, lockh, sizeof(plock));
+        LUSTRE_IT(it)->it_lock_mode = 0;
+        LUSTRE_IT(it)->it_data = NULL;
+        LASSERT((body->valid & OBD_MD_FID) != 0);
+
+        memcpy(&rdata, data, sizeof(rdata));
+        rdata.id1 = body->id1;
+        rdata.name = NULL;
+        rdata.namelen = 0;
+
+        LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE;
+        ptlrpc_req_finished(req);
+
+        rc = md_enqueue(lmv->tgts[id_group(&rdata.id1)].ltd_exp, 
+                        lock_type, it, lock_mode, &rdata, lockh, lmm, 
+                        lmmsize, cb_compl, cb_blocking, cb_data);
+        ldlm_lock_decref(&plock, pmode);
         RETURN(rc);
 }
 
@@ -1038,11 +1117,15 @@ int lmv_enqueue(struct obd_export *exp, int lock_type,
         rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp, 
                         lock_type, it, lock_mode, data, lockh, lmm, 
                         lmmsize, cb_compl, cb_blocking, cb_data);
+        if (rc == 0 && it->it_op == IT_OPEN)
+                rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode,
+                                        data, lockh, lmm, lmmsize,
+                                        cb_compl, cb_blocking, cb_data);
         RETURN(rc);
 }
 
 int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
-                     char *filename, int namelen, unsigned long valid,
+                     char *filename, int namelen, __u64 valid,
                      unsigned int ea_size, struct ptlrpc_request **request)
 {
         int rc, mds = id_group(id), loop = 0;
@@ -1051,7 +1134,6 @@ int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
         struct lustre_id rid = *id;
         struct mds_body *body;
         struct lmv_obj *obj;
-        int old_valid;
         ENTRY;
         
         rc = lmv_check_connect(obd);
@@ -1071,18 +1153,9 @@ repeat:
         CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
                namelen, filename, OLID4(id), OLID4(&rid));
 
-       old_valid = valid;
-
-        /*
-         * here should be applied OBD_MD_FID to ->valid, because otherwise,
-         * mds_getattr_lock() will not fetch fid component of lustre_id and
-         * thus, next call to md_getattr_lock() will be performed to wrong mds.
-         */
-        if (!(old_valid & OBD_MD_FID))
-                valid |= OBD_MD_FID;
-        
-        rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp, 
-                             &rid, filename, namelen, valid,
+        rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
+                             &rid, filename, namelen,
+                             valid == OBD_MD_FLID ? valid : valid | OBD_MD_FID,
                              ea_size, request);
         if (rc == 0) {
                 /*
@@ -1092,20 +1165,15 @@ repeat:
                  */
                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
                 LASSERT(body != NULL);
+                LASSERT((body->valid & OBD_MD_FID) != 0
+                                || body->valid == OBD_MD_FLID);
+
                 if (body->valid & OBD_MD_MDS) {
                         struct ptlrpc_request *req = NULL;
                         
                         rid = body->id1;
                         CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
 
-                        /* 
-                         * turning OBD_MD_FID fetching off, as we already have
-                         * full lustre_id and do need to fetch fid component
-                         * again. This will help to make thing slightly faster.
-                         */
-                        if (!(old_valid & OBD_MD_FID))
-                                valid &= ~OBD_MD_FID;
-                        
                         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp, 
                                              &rid, NULL, 1, valid, ea_size, &req);
                         ptlrpc_req_finished(*request);
@@ -1175,7 +1243,8 @@ int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
         ENTRY;
 
         CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
-               oldlen, old, OLID4(&data->id1), newlen, new, OLID4(&data->id2));
+               oldlen, old, OLID4(&data->id1), newlen, new,
+               OLID4(&data->id2));
 
         rc = lmv_check_connect(obd);
        if (rc)
@@ -1302,6 +1371,7 @@ int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
                                               sizeof(*body));
                         LASSERT(body != NULL);
+                        LASSERT((body->valid & OBD_MD_FID) != 0);
                         LASSERT(id_group(&body->id1) == id_group(&data->id1));
                 }
         }
@@ -1365,10 +1435,10 @@ int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
 
 void lmv_remove_dots(struct page *page)
 {
-        char *kaddr = page_address(page);
         unsigned limit = PAGE_CACHE_SIZE;
-        unsigned offs, rec_len;
+        char *kaddr = page_address(page);
         struct ext2_dir_entry_2 *p;
+        unsigned offs, rec_len;
 
         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
@@ -1436,38 +1506,49 @@ int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
         struct obd_device *obd = exp->exp_obd;
         struct lmv_obd *lmv = &obd->u.lmv;
         struct mea *mea = data->mea1;
-        struct mdc_op_data data2;
-        int i, rc = 0, mds;
+        struct mdc_op_data *data2;
+        int i, rc = 0;
         ENTRY;
 
+        OBD_ALLOC(data2, sizeof(*data2));
+        if (data2 == NULL)
+                RETURN(-ENOMEM);
+        
         LASSERT(mea != NULL);
         for (i = 0; i < mea->mea_count; i++) {
-                memset(&data2, 0, sizeof(data2));
-                data2.id1 = mea->mea_ids[i];
-                data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
+                memset(data2, 0, sizeof(*data2));
+                data2->id1 = mea->mea_ids[i];
+                data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
                 
-                mds = id_group(&data2.id1);
-
-                if (lmv->tgts[mds].ltd_exp == NULL)
+                if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
                         continue;
 
-                rc = md_unlink(lmv->tgts[mds].ltd_exp, &data2, req);
+                rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
+                               data2, req);
+                
                 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
                        OLID4(&mea->mea_ids[i]), rc);
+                
                 if (*req) {
                         ptlrpc_req_finished(*req);
                         *req = NULL;
                 }
                 if (rc)
-                        break;
+                        RETURN(rc);
         }
+        OBD_FREE(data2, sizeof(*data2));
         RETURN(rc);
 }
 
-int lmv_put_inode(struct obd_export *exp, struct lustre_id *id)
+int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
 {
         ENTRY;
-        lmv_delete_obj(exp, id);
+
+        LASSERT(exp && id);
+        if (lmv_delete_obj(exp, id)) {
+                CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
+                       OLID4(id));
+        }
         RETURN(0);
 }
 
@@ -1487,7 +1568,9 @@ int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
                 /* mds asks to remove slave objects */
                 rc = lmv_unlink_slaves(exp, data, request);
                 RETURN(rc);
-        } else if (data->namelen != 0) {
+        }
+
+        if (data->namelen != 0) {
                 struct lmv_obj *obj;
                 
                 obj = lmv_grab_obj(obd, &data->id1);
@@ -1505,12 +1588,12 @@ int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
                        OLID4(&data->id1));
         }
         rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp, 
-                       data, request); 
+                       data, request);
         RETURN(rc);
 }
 
 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
-                                    char *name, int len)
+                                    struct lustre_id *id)
 {
         struct obd_device *obd = exp->exp_obd;
         struct lmv_obd *lmv = &obd->u.lmv;
@@ -1520,12 +1603,14 @@ struct obd_device *lmv_get_real_obd(struct obd_export *exp,
         rc = lmv_check_connect(obd);
        if (rc)
                RETURN(ERR_PTR(rc));
-        obd = lmv->tgts[0].ltd_exp->exp_obd;
+        obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd;
         EXIT;
+        
         return obd;
 }
 
-int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
+int lmv_init_ea_size(struct obd_export *exp, int easize,
+                     int cookiesize)
 {
         struct obd_device *obd = exp->exp_obd;
         struct lmv_obd *lmv = &obd->u.lmv;
@@ -1563,6 +1648,7 @@ int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
 }
 
 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
+                          void *acl, int acl_size,
                           struct lov_stripe_md **ea, struct obd_trans_info *oti)
 {
         struct obd_device *obd = exp->exp_obd;
@@ -1575,7 +1661,8 @@ int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
         LASSERT(ea == NULL);
         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
 
-        rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa, &obj_mdp, oti);
+        rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa,
+                        acl, acl_size, &obj_mdp, oti);
 
         RETURN(rc);
 }
@@ -1595,6 +1682,7 @@ int lmv_getready(struct obd_export *exp)
  * values for "master" object, as it will be used.
  */
 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
+                   void *acl, int acl_size,
                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
 {
         struct obd_device *obd = exp->exp_obd;
@@ -1611,12 +1699,15 @@ int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
         LASSERT(oa != NULL);
         
         if (ea == NULL) {
-                rc = lmv_obd_create_single(exp, oa, NULL, oti);
+                rc = lmv_obd_create_single(exp, oa, acl, acl_size, NULL, oti);
                 if (rc)
                         CERROR("Can't create object, rc = %d\n", rc);
                 RETURN(rc);
         }
 
+        /* acl is only suppied when mds create single remote obj */
+        LASSERT(acl == NULL && acl_size == 0);
+
         if (*ea == NULL) {
                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
                 if (rc < 0) {
@@ -1669,7 +1760,8 @@ int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
                         OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
 
-                rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
+                rc = obd_create(lmv->tgts[c].ltd_exp, oa, NULL, 0,
+                                &obj_mdp, oti);
                 if (rc) {
                         CERROR("obd_create() failed on MDT target %d, "
                                "error %d\n", c, rc);
@@ -1777,6 +1869,26 @@ static int lmv_get_info(struct obd_export *exp, __u32 keylen,
                 struct lmv_desc *desc_ret = val;
                 *desc_ret = lmv->desc;
                 RETURN(0);
+        } else if (keylen == strlen("remote_flag") &&
+                   !strcmp(key, "remote_flag")) {
+                struct lmv_tgt_desc *tgts;
+                int i;
+
+                LASSERT(*vallen == sizeof(__u32));
+                for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
+                     i++, tgts++) {
+
+                        /* all tgts should be connected when this get called. */
+                        if (!tgts || !tgts->ltd_exp) {
+                                CERROR("target not setup?\n");
+                                continue;
+                        }
+
+                        if (!obd_get_info(tgts->ltd_exp, keylen, key,
+                                          vallen, val))
+                                RETURN(0);
+                }
+                RETURN(-EINVAL);
         }
 
         CDEBUG(D_IOCTL, "invalid key\n");
@@ -1803,7 +1915,46 @@ int lmv_set_info(struct obd_export *exp, obd_count keylen,
                 lmv_set_timeouts(obd);
                 RETURN(0);
         }
-        
+
+        /* maybe this could be default */
+        if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) ||
+            (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) {
+                struct lmv_tgt_desc *tgt;
+                struct obd_export *exp;
+                int rc = 0, err, i;
+
+                spin_lock(&lmv->lmv_lock);
+                for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
+                     i++, tgt++) {
+                        exp = tgt->ltd_exp;
+                        /* during setup time the connections to mdc might
+                         * haven't been established.
+                         */
+                        if (exp == NULL) {
+                                struct obd_device *tgt_obd;
+
+                                tgt_obd = class_find_client_obd(&tgt->uuid,
+                                                                LUSTRE_MDC_NAME,
+                                                                &obd->obd_uuid);
+                                if (!tgt_obd) {
+                                        CERROR("can't set info %s, "
+                                               "device %s not attached?\n",
+                                                (char *) key, tgt->uuid.uuid);
+                                        rc = -EINVAL;
+                                        continue;
+                                }
+                                exp = tgt_obd->obd_self_export;
+                        }
+
+                        err = obd_set_info(exp, keylen, key, vallen, val);
+                        if (!rc)
+                                rc = err;
+                }
+                spin_unlock(&lmv->lmv_lock);
+
+                RETURN(rc);
+        }
+
         RETURN(-EINVAL);
 }
 
@@ -1812,11 +1963,12 @@ int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
 {
         struct obd_device *obd = class_exp2obd(exp);
         struct lmv_obd *lmv = &obd->u.lmv;
-        int mea_size;
+        struct mea *meap, *lsmp;
+        int mea_size, i;
         ENTRY;
 
-       mea_size = sizeof(struct lustre_id) * 
-                lmv->desc.ld_tgt_count + sizeof(struct mea);
+       mea_size = (sizeof(struct lustre_id) * 
+                    lmv->desc.ld_tgt_count) + sizeof(struct mea);
         if (!lmmp)
                 RETURN(mea_size);
 
@@ -1835,43 +1987,88 @@ int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
         if (!lsm)
                 RETURN(mea_size);
 
-#warning "MEA packing/convertation must be here! -bzzz"
-        memcpy(*lmmp, lsm, mea_size);
+        lsmp = (struct mea *)lsm;
+        meap = (struct mea *)*lmmp;
+
+        if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
+            lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
+                RETURN(-EINVAL);
+
+        meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
+        meap->mea_count = cpu_to_le32(lsmp->mea_count);
+        meap->mea_master = cpu_to_le32(lsmp->mea_master);
+
+        for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+                meap->mea_ids[i] = meap->mea_ids[i];
+                id_cpu_to_le(&meap->mea_ids[i]);
+        }
+
         RETURN(mea_size);
 }
 
-int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
-                 struct lov_mds_md *disk_src, int mdsize)
+int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
+                 struct lov_mds_md *lmm, int lmm_size)
 {
         struct obd_device *obd = class_exp2obd(exp);
+        struct mea **tmea = (struct mea **)lsmp;
+        struct mea *mea = (struct mea *)lmm;
         struct lmv_obd *lmv = &obd->u.lmv;
-        struct mea **tmea = (struct mea **) mem_tgt;
-        struct mea *mea = (void *) disk_src;
-        int mea_size;
+        int mea_size, i, rc = 0;
+        __u32 magic;
         ENTRY;
 
-       mea_size = sizeof(struct lustre_id) * 
+        mea_size = sizeof(struct lustre_id) * 
                 lmv->desc.ld_tgt_count + sizeof(struct mea);
-        if (mem_tgt == NULL)
+
+        if (lsmp == NULL)
                 return mea_size;
 
-        if (*mem_tgt != NULL && disk_src == NULL) {
+        if (*lsmp != NULL && lmm == NULL) {
                 OBD_FREE(*tmea, mea_size);
                 RETURN(0);
         }
 
-        LASSERT(mea_size == mdsize);
+        LASSERT(mea_size == lmm_size);
 
         OBD_ALLOC(*tmea, mea_size);
         if (*tmea == NULL)
                 RETURN(-ENOMEM);
 
-        if (!disk_src)
+        if (!lmm)
                 RETURN(mea_size);
 
-#warning "MEA unpacking/convertation must be here! -bzzz"
-        memcpy(*tmea, mea, mdsize);
+        if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
+            mea->mea_magic == MEA_MAGIC_ALL_CHARS)
+        {
+                magic = le32_to_cpu(mea->mea_magic);
+        } else {
+                struct mea_old *old = (struct mea_old *)lmm;
+        
+                mea_size = sizeof(struct lustre_id) * old->mea_count + 
+                        sizeof(struct mea_old);
+        
+                if (old->mea_count > 256 || old->mea_master > 256 ||
+                    lmm_size < mea_size || old->mea_master > old->mea_count) {
+                        CWARN("bad MEA: count %u, master %u, size %u\n",
+                              old->mea_count, old->mea_master, mea_size);
+                        GOTO(out_free_mea, rc = -EINVAL);
+                }
+                magic = MEA_MAGIC_LAST_CHAR;
+        }
+
+        (*tmea)->mea_magic = magic;
+        (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
+        (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
+
+        for (i = 0; i < (*tmea)->mea_count; i++) {
+                (*tmea)->mea_ids[i] = mea->mea_ids[i];
+                id_le_to_cpu(&(*tmea)->mea_ids[i]);
+        }
         RETURN(mea_size);
+
+out_free_mea:
+        OBD_FREE(*tmea, mea_size);
+        return rc;
 }
 
 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
@@ -1939,7 +2136,7 @@ struct md_ops lmv_md_ops = {
         .m_unlink              = lmv_unlink,
         .m_get_real_obd        = lmv_get_real_obd,
         .m_valid_attrs         = lmv_valid_attrs,
-        .m_put_inode           = lmv_put_inode,
+        .m_delete_inode        = lmv_delete_inode,
 };
 
 int __init lmv_init(void)
@@ -1959,6 +2156,9 @@ int __init lmv_init(void)
         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
                                  lvars.module_vars,
                                  OBD_LMV_DEVICENAME);
+        if (rc)
+                kmem_cache_destroy(obj_cache);
+        
         return rc;
 }