Whamcloud - gitweb
add support for build HEAD without vfs_intent patches.
[fs/lustre-release.git] / lustre / lmv / lmv_intent.c
index 205d4a7..206cb11 100644 (file)
@@ -1,7 +1,7 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
+ * Copyright (C) 2002, 2003, 2004, 2005, 2006 Cluster File Systems, Inc.
  *
  *   This file is part of Lustre, http://www.lustre.org.
  *
 #include <linux/pagemap.h>
 #include <asm/div64.h>
 #include <linux/seq_file.h>
+#include <linux/namei.h>
+# ifndef HAVE_VFS_INTENT_PATCHES
+# include <linux/lustre_intent.h>
+# endif
 #else
 #include <liblustre.h>
 #endif
 
-#include <linux/obd_support.h>
-#include <linux/lustre_lib.h>
-#include <linux/lustre_net.h>
-#include <linux/lustre_idl.h>
-#include <linux/lustre_dlm.h>
-#include <linux/lustre_mds.h>
-#include <linux/obd_class.h>
-#include <linux/obd_ost.h>
-#include <linux/lprocfs_status.h>
-#include <linux/lustre_fsfilt.h>
-#include <linux/obd_lmv.h>
-#include <linux/namei.h>
-#include <linux/lustre_lite.h>
+#include <lustre/lustre_idl.h>
+#include <obd_support.h>
+#include <lustre_lib.h>
+#include <lustre_net.h>
+#include <lustre_dlm.h>
+#include <obd_class.h>
+#include <lprocfs_status.h>
 #include "lmv_internal.h"
 
-
 static inline void lmv_drop_intent_lock(struct lookup_intent *it)
 {
-        if (LUSTRE_IT(it)->it_lock_mode != 0)
-                ldlm_lock_decref((void *)&LUSTRE_IT(it)->it_lock_handle,
-                                 LUSTRE_IT(it)->it_lock_mode);
+        if (it->d.lustre.it_lock_mode != 0) {
+                ldlm_lock_decref((void *)&it->d.lustre.it_lock_handle,
+                                 it->d.lustre.it_lock_mode);
+                it->d.lustre.it_lock_mode = 0;
+        }
 }
 
-int lmv_handle_remote_inode(struct obd_export *exp, void *lmm,
-                            int lmmsize, struct lookup_intent *it,
-                            int flags, struct ptlrpc_request **reqp,
-                            ldlm_blocking_callback cb_blocking)
+int lmv_intent_remote(struct obd_export *exp, void *lmm,
+                      int lmmsize, struct lookup_intent *it,
+                      int flags, struct ptlrpc_request **reqp,
+                      ldlm_blocking_callback cb_blocking,
+                      int extra_lock_flags)
 {
         struct obd_device *obd = exp->exp_obd;
         struct lmv_obd *lmv = &obd->u.lmv;
-        struct mds_body *body = NULL;
-        int rc = 0;
+        struct ptlrpc_request *req = NULL;
+        struct lustre_handle plock;
+        struct md_op_data *op_data;
+        struct obd_export *tgt_exp;
+        struct mdt_body *body;
+        int pmode, rc = 0;
         ENTRY;
 
-        body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
+        body = lustre_msg_buf((*reqp)->rq_repmsg,
+                              DLM_REPLY_REC_OFF, sizeof(*body));
         LASSERT(body != NULL);
+        LASSERT(lustre_rep_swabbed(*reqp, DLM_REPLY_REC_OFF));
 
-        if (body->valid & OBD_MD_MDS) {
+        if (!(body->valid & OBD_MD_MDS))
+                RETURN(0);
+
+        /*
+         * oh, MDS reports that this is remote inode case i.e. we have to ask
+         * for real attrs on another MDS.
+         */
+        if (it->it_op & IT_LOOKUP) {
                 /*
-                 * oh, MDS reports that this is remote inode case i.e. we have
-                 * to ask for real attrs on another MDS.
+                 * unfortunately, we have to lie to MDC/MDS to retrieve
+                 * attributes llite needs.
                  */
-                struct ptlrpc_request *req = NULL;
-                struct lustre_handle plock;
-                struct lustre_id nid;
-                int pmode;
+                it->it_op = IT_GETATTR;
+        }
 
-                if (it->it_op == IT_LOOKUP || it->it_op == IT_CHDIR) {
-                        /*
-                         * unfortunately, we have to lie to MDC/MDS to retrieve
-                         * attributes llite needs.
-                         */
-                        it->it_op = IT_GETATTR;
-                }
+        /* we got LOOKUP lock, but we really need attrs */
+        pmode = it->d.lustre.it_lock_mode;
+        if (pmode) {
+                plock.cookie = it->d.lustre.it_lock_handle;
+                it->d.lustre.it_lock_mode = 0;
+                it->d.lustre.it_data = 0;
+        }
 
-                /* we got LOOKUP lock, but we really need attrs */
-                pmode = LUSTRE_IT(it)->it_lock_mode;
-                if (pmode) {
-                        memcpy(&plock, &LUSTRE_IT(it)->it_lock_handle,
-                               sizeof(plock));
-                        LUSTRE_IT(it)->it_lock_mode = 0;
-                }
+        LASSERT(fid_is_sane(&body->fid1));
 
-                LASSERT((body->valid & OBD_MD_FID) != 0);
-                
-                nid = body->id1;
-                LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE;
-                rc = md_intent_lock(lmv->tgts[id_group(&nid)].ltd_exp, &nid, NULL,
-                                    0, lmm, lmmsize, NULL, it, flags, &req, cb_blocking);
+        it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
 
-                /*
-                 * llite needs LOOKUP lock to track dentry revocation in order
-                 * to maintain dcache consistency. Thus drop UPDATE lock here
-                 * and put LOOKUP in request.
-                 */
-                if (rc == 0) {
-                        lmv_drop_intent_lock(it);
-                        memcpy(&LUSTRE_IT(it)->it_lock_handle, &plock,
-                               sizeof(plock));
-                        LUSTRE_IT(it)->it_lock_mode = pmode;
-                } else if (pmode)
-                        ldlm_lock_decref(&plock, pmode);
+        tgt_exp = lmv_find_export(lmv, &body->fid1);
+        if (IS_ERR(tgt_exp))
+                GOTO(out, rc = PTR_ERR(tgt_exp));
 
-                ptlrpc_req_finished(*reqp);
-                *reqp = req;
+        OBD_ALLOC_PTR(op_data);
+        if (op_data == NULL)
+                GOTO(out, rc = -ENOMEM);
+
+        op_data->op_fid1 = body->fid1;
+        op_data->op_bias = MDS_CROSS_REF;
+
+        rc = md_intent_lock(tgt_exp, op_data, lmm, lmmsize, it, flags,
+                            &req, cb_blocking, extra_lock_flags);
+
+        /*
+         * llite needs LOOKUP lock to track dentry revocation in order to
+         * maintain dcache consistency. Thus drop UPDATE lock here and put
+         * LOOKUP in request.
+         */
+        if (rc == 0) {
+                lmv_drop_intent_lock(it);
+                it->d.lustre.it_lock_handle = plock.cookie;
+                it->d.lustre.it_lock_mode = pmode;
+        }
+
+        OBD_FREE_PTR(op_data);
+        EXIT;
+out:
+        if (rc && pmode)
+                ldlm_lock_decref(&plock, pmode);
+
+        ptlrpc_req_finished(*reqp);
+        *reqp = req;
+        return rc;
+}
+
+int lmv_alloc_slave_fids(struct obd_device *obd, struct lu_fid *pid,
+                         struct md_op_data *op, struct lu_fid *fid)
+{
+        struct lmv_obd *lmv = &obd->u.lmv;
+        struct lmv_obj *obj;
+        mdsno_t mds;
+        int mea_idx;
+        int rc;
+        ENTRY;
+
+        obj = lmv_obj_grab(obd, pid);
+        if (!obj) {
+                CERROR("Object "DFID" should be split\n",
+                       PFID(pid));
+                RETURN(0);
+        }
+
+        mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
+                               (char *)op->op_name, op->op_namelen);
+        mds = obj->lo_inodes[mea_idx].li_mds;
+        lmv_obj_put(obj);
+
+        rc = __lmv_fid_alloc(lmv, fid, mds);
+        if (rc) {
+                CERROR("Can't allocate new fid, rc %d\n",
+                       rc);
+                RETURN(rc);
         }
+
+        CDEBUG(D_INFO, "Allocate new fid "DFID" for split "
+               "obj\n", PFID(fid));
+
         RETURN(rc);
 }
 
-int lmv_intent_open(struct obd_export *exp, struct lustre_id *pid,
-                    const char *name, int len, void *lmm, int lmmsize,
-                    struct lustre_id *cid, struct lookup_intent *it,
+/*
+ * IT_OPEN is intended to open (and create, possible) an object. Parent (pid)
+ * may be split dir.
+ */
+int lmv_intent_open(struct obd_export *exp, struct md_op_data *op_data,
+                    void *lmm, int lmmsize, struct lookup_intent *it,
                     int flags, struct ptlrpc_request **reqp,
-                    ldlm_blocking_callback cb_blocking)
+                    ldlm_blocking_callback cb_blocking,
+                    int extra_lock_flags)
 {
         struct obd_device *obd = exp->exp_obd;
+        struct lu_fid rpid = op_data->op_fid1;
         struct lmv_obd *lmv = &obd->u.lmv;
-        struct mds_body *body = NULL;
-        struct lustre_id rpid = *pid;
-        int rc, mds, loop = 0;
+        struct md_op_data *sop_data;
+        struct obd_export *tgt_exp;
+        struct lmv_stripe_md *mea;
+        struct mdt_body *body;
         struct lmv_obj *obj;
-        struct mea *mea;
+        int rc, loop = 0;
         ENTRY;
 
-        /* IT_OPEN is intended to open (and create, possible) an object. Parent
-         * (pid) may be splitted dir */
+        OBD_ALLOC_PTR(sop_data);
+        if (sop_data == NULL)
+                RETURN(-ENOMEM);
+
+        /* save op_data fro repeat case */
+        *sop_data = *op_data;
 
 repeat:
-        LASSERT(++loop <= 2);
-        mds = id_group(&rpid);
-        obj = lmv_grab_obj(obd, &rpid);
+
+        ++loop;
+        LASSERT(loop <= 2);
+        obj = lmv_obj_grab(obd, &rpid);
         if (obj) {
-                /* directory is already splitted, so we have to forward
-                 * request to the right MDS */
-                mds = raw_name2idx(obj->hashtype, obj->objcount, 
-                                   (char *)name, len);
-                
-                CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n",
-                       mds, OLID4(&rpid));
-                rpid = obj->objs[mds].id;
-                lmv_put_obj(obj);
+                int mea_idx;
+
+                /*
+                 * Directory is already split, so we have to forward request to
+                 * the right MDS.
+                 */
+                mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
+                                       (char *)op_data->op_name,
+                                       op_data->op_namelen);
+
+                rpid = obj->lo_inodes[mea_idx].li_fid;
+
+                sop_data->op_mds = obj->lo_inodes[mea_idx].li_mds;
+                tgt_exp = lmv_get_export(lmv, sop_data->op_mds);
+                sop_data->op_bias &= ~MDS_CHECK_SPLIT;
+                lmv_obj_put(obj);
+                CDEBUG(D_OTHER, "Choose slave dir ("DFID")\n", PFID(&rpid));
+        } else {
+                struct lmv_tgt_desc *tgt;
+
+                sop_data->op_bias |= MDS_CHECK_SPLIT;
+                tgt = lmv_find_target(lmv, &rpid);
+                sop_data->op_mds = tgt->ltd_idx;
+                tgt_exp = tgt->ltd_exp;
+        }
+        if (IS_ERR(tgt_exp))
+                GOTO(out_free_sop_data, rc = PTR_ERR(tgt_exp));
+
+        sop_data->op_fid1 = rpid;
+
+        if (it->it_op & IT_CREAT) {
+                /*
+                 * For open with IT_CREATE and for IT_CREATE cases allocate new
+                 * fid and setup FLD for it.
+                 */
+                rc = lmv_fid_alloc(exp, &sop_data->op_fid2, sop_data);
+                if (rc)
+                        GOTO(out_free_sop_data, rc);
+
+                if (rc == -ERESTART)
+                        goto repeat;
+                else if (rc)
+                        GOTO(out_free_sop_data, rc);
         }
 
-        rc = md_intent_lock(lmv->tgts[id_group(&rpid)].ltd_exp, &rpid, name,
-                            len, lmm, lmmsize, cid, it, flags, reqp, cb_blocking);
+        rc = md_intent_lock(tgt_exp, sop_data, lmm, lmmsize, it, flags,
+                            reqp, cb_blocking, extra_lock_flags);
+
         if (rc == -ERESTART) {
-                /* directory got splitted. time to update local object and
-                 * repeat the request with proper MDS */
-                LASSERT(id_equal_fid(pid, &rpid));
-                rc = lmv_get_mea_and_update_object(exp, &rpid);
+                LASSERT(*reqp != NULL);
+                DEBUG_REQ(D_WARNING|D_RPCTRACE, *reqp,
+                          "Got -ERESTART during open!\n");
+                ptlrpc_req_finished(*reqp);
+                *reqp = NULL;
+                it->d.lustre.it_data = 0;
+
+                /*
+                 * Directory got split. Time to update local object and repeat
+                 * the request with proper MDS.
+                 */
+                LASSERT(lu_fid_eq(&op_data->op_fid1, &rpid));
+                rc = lmv_handle_split(exp, &rpid);
                 if (rc == 0) {
-                        ptlrpc_req_finished(*reqp);
-                        goto repeat;
+                        /* We should reallocate child FID. */
+                        rc = lmv_alloc_slave_fids(obd, &rpid, op_data,
+                                                  &sop_data->op_fid2);
+                        if (rc == 0)
+                                goto repeat;
                 }
         }
+
         if (rc != 0)
-                RETURN(rc);
+                GOTO(out_free_sop_data, rc);
 
-        /* okay, MDS has returned success. Probably name has been resolved in
-         * remote inode */
-        rc = lmv_handle_remote_inode(exp, lmm, lmmsize, it,
-                                     flags, reqp, cb_blocking);
+        /*
+         * Okay, MDS has returned success. Probably name has been resolved in
+         * remote inode.
+         */
+        rc = lmv_intent_remote(exp, lmm, lmmsize, it, flags, reqp,
+                               cb_blocking, extra_lock_flags);
         if (rc != 0) {
                 LASSERT(rc < 0);
-
-                /* 
-                 * this is possible, that some userspace application will try to
-                 * open file as directory and we will have error -20 here. As
-                 * this is "usual" situation, we should not print error here,
+                /*
+                 * This is possible, that some userspace application will try to
+                 * open file as directory and we will have -ENOTDIR here. As
+                 * this is normal situation, we should not print error here,
                  * only debug info.
                  */
-                CDEBUG(D_OTHER, "can't handle remote %s: dir "DLID4"("DLID4"):"
-                       "%*s: %d\n", LL_IT2STR(it), OLID4(pid), OLID4(&rpid),
-                       len, name, rc);
-                RETURN(rc);
+                CDEBUG(D_OTHER, "can't handle remote %s: dir "DFID"("DFID"):"
+                       "%*s: %d\n", LL_IT2STR(it), PFID(&op_data->op_fid2),
+                       PFID(&rpid), op_data->op_namelen, op_data->op_name, rc);
+                GOTO(out_free_sop_data, rc);
         }
 
         /*
-         * nothing is found, do not access body->id1 as it is zero and thus
+         * Nothing is found, do not access body->fid1 as it is zero and thus
          * pointless.
          */
-        if (LUSTRE_IT(it)->it_disposition & DISP_LOOKUP_NEG)
-                RETURN(0);
+        if ((it->d.lustre.it_disposition & DISP_LOOKUP_NEG) &&
+            !(it->d.lustre.it_disposition & DISP_OPEN_CREATE) &&
+            !(it->d.lustre.it_disposition & DISP_OPEN_OPEN))
+                GOTO(out_free_sop_data, rc = 0);
 
         /* caller may use attrs MDS returns on IT_OPEN lock request so, we have
-         * to update them for splitted dir */
-        body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
+         * to update them for split dir */
+        body = lustre_msg_buf((*reqp)->rq_repmsg,
+                              DLM_REPLY_REC_OFF, sizeof(*body));
         LASSERT(body != NULL);
-        LASSERT((body->valid & OBD_MD_FID) != 0);
-        
-        cid = &body->id1;
-        obj = lmv_grab_obj(obd, cid);
-        if (!obj && (mea = lmv_splitted_dir_body(*reqp, 1))) {
-                /* wow! this is splitted dir, we'd like to handle it */
-                obj = lmv_create_obj(exp, &body->id1, mea);
+        LASSERT(lustre_rep_swabbed(*reqp, DLM_REPLY_REC_OFF));
+
+        /* could not find object, FID is not present in response. */
+        if (!(body->valid & OBD_MD_FLID))
+                GOTO(out_free_sop_data, rc = 0);
+
+        obj = lmv_obj_grab(obd, &body->fid1);
+        if (!obj && (mea = lmv_get_mea(*reqp, DLM_REPLY_REC_OFF))) {
+                /* FIXME: capability for remote! */
+                /* wow! this is split dir, we'd like to handle it */
+                obj = lmv_obj_create(exp, &body->fid1, mea);
                 if (IS_ERR(obj))
-                        RETURN(PTR_ERR(obj));
+                        GOTO(out_free_sop_data, rc = (int)PTR_ERR(obj));
         }
 
         if (obj) {
-                /* this is splitted dir and we'd want to get attrs */
-                CDEBUG(D_OTHER, "attrs from slaves for "DLID4"\n",
-                       OLID4(cid));
-                
-                rc = lmv_revalidate_slaves(exp, reqp, cid, it, 1,
-                                           cb_blocking);
+                /* This is split dir and we'd want to get attrs. */
+                CDEBUG(D_OTHER, "attrs from slaves for "DFID"\n",
+                       PFID(&body->fid1));
+
+                rc = lmv_revalidate_slaves(exp, reqp, &body->fid1, it, 1,
+                                           cb_blocking, extra_lock_flags);
         } else if (S_ISDIR(body->mode)) {
-                CDEBUG(D_OTHER, "object "DLID4" has not lmv obj?\n",
-                       OLID4(cid));
+                CDEBUG(D_OTHER, "object "DFID" has not lmv obj?\n",
+                       PFID(&body->fid1));
         }
-        
+
         if (obj)
-                lmv_put_obj(obj);
-        
-        RETURN(rc);
+                lmv_obj_put(obj);
+
+        EXIT;
+out_free_sop_data:
+        OBD_FREE_PTR(sop_data);
+        return rc;
 }
 
-int lmv_intent_getattr(struct obd_export *exp, struct lustre_id *pid,
-                       const char *name, int len, void *lmm, int lmmsize,
-                       struct lustre_id *cid, struct lookup_intent *it,
+int lmv_intent_getattr(struct obd_export *exp, struct md_op_data *op_data,
+                       void *lmm, int lmmsize, struct lookup_intent *it,
                        int flags, struct ptlrpc_request **reqp,
-                       ldlm_blocking_callback cb_blocking)
+                       ldlm_blocking_callback cb_blocking,
+                       int extra_lock_flags)
 {
+        struct lmv_obj *obj = NULL, *obj2 = NULL;
         struct obd_device *obd = exp->exp_obd;
+        struct lu_fid rpid = op_data->op_fid1;
         struct lmv_obd *lmv = &obd->u.lmv;
-        struct mds_body *body = NULL;
-        struct lustre_id rpid = *pid;
-        struct lmv_obj *obj = NULL, *obj2 = NULL;
-        struct mea *mea;
-        int rc = 0, mds;
+        struct md_op_data *sop_data;
+        struct lmv_stripe_md *mea;
+        struct mdt_body *body;
+        mdsno_t mds;
+        int rc = 0;
         ENTRY;
 
-        if (cid) {
-                /* caller wants to revalidate attrs of obj we have to revalidate
-                 * slaves if requested object is splitted directory */
-                CDEBUG(D_OTHER, "revalidate attrs for "DLID4"\n", OLID4(cid));
-                mds = id_group(cid);
+        OBD_ALLOC_PTR(sop_data);
+        if (sop_data == NULL)
+                RETURN(-ENOMEM);
+
+        /* save op_data fro repeat case */
+        *sop_data = *op_data;
+
+        if (fid_is_sane(&op_data->op_fid2)) {
+                /*
+                 * Caller wants to revalidate attrs of obj we have to revalidate
+                 * slaves if requested object is split directory.
+                 */
+                CDEBUG(D_OTHER, "revalidate attrs for "DFID"\n",
+                       PFID(&op_data->op_fid2));
+
+                rc = lmv_fld_lookup(lmv, &op_data->op_fid2, &mds);
+                if (rc)
+                        GOTO(out_free_sop_data, rc);
 #if 0
-                obj = lmv_grab_obj(obd, cid);
+                /*
+                 * In fact, we do not need this with current intent_lock(), but
+                 * it may change some day.
+                 */
+                obj = lmv_obj_grab(obd, &op_data->op_fid2);
                 if (obj) {
-                        /* in fact, we need not this with current intent_lock(),
-                         * but it may change some day */
-                        if (!id_equal_fid(pid, cid)){
-                                rpid = obj->objs[mds].id;
-                                mds = id_group(&rpid);
+                        if (!lu_fid_eq(&op_data->op_fid1, &op_data->op_fid2)){
+                                rpid = obj->lo_inodes[mds].li_fid;
+                                mds = obj->lo_inodes[mds].li_mds;
                         }
-                        lmv_put_obj(obj);
+                        lmv_obj_put(obj);
                 }
 #endif
         } else {
-                CDEBUG(D_OTHER, "INTENT getattr for %*s on "DLID4"\n",
-                       len, name, OLID4(pid));
-                mds = id_group(pid);
-                obj = lmv_grab_obj(obd, pid);
-                if (obj && len) {
-                        /* directory is already splitted. calculate mds */
-                        mds = raw_name2idx(obj->hashtype, obj->objcount, 
-                                           (char *)name, len);
-                        rpid = obj->objs[mds].id;
-                        mds = id_group(&rpid);
-                        lmv_put_obj(obj);
-
-                        CDEBUG(D_OTHER, "forward to MDS #%u (slave "DLID4")\n",
-                               mds, OLID4(&rpid));
+                CDEBUG(D_OTHER, "INTENT getattr for %*s on "DFID"\n",
+                       op_data->op_namelen, op_data->op_name,
+                       PFID(&op_data->op_fid1));
+
+                rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds);
+                if (rc)
+                        GOTO(out_free_sop_data, rc);
+                obj = lmv_obj_grab(obd, &op_data->op_fid1);
+                if (obj && op_data->op_namelen) {
+                        int mea_idx;
+
+                        /* directory is already split. calculate mds */
+                        mea_idx = raw_name2idx(obj->lo_hashtype, obj->lo_objcount,
+                                               (char *)op_data->op_name,
+                                               op_data->op_namelen);
+                        rpid = obj->lo_inodes[mea_idx].li_fid;
+                        mds = obj->lo_inodes[mea_idx].li_mds;
+                        sop_data->op_bias &= ~MDS_CHECK_SPLIT;
+                        lmv_obj_put(obj);
+
+                        CDEBUG(D_OTHER, "forward to MDS #"LPU64" (slave "DFID")\n",
+                               mds, PFID(&rpid));
+                } else {
+                        rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds);
+                        if (rc)
+                                GOTO(out_free_sop_data, rc);
+                        sop_data->op_bias |= MDS_CHECK_SPLIT;
                 }
         }
 
-        /* the same about fid returning. */
-        rc = md_intent_lock(lmv->tgts[mds].ltd_exp, &rpid, name, len, lmm,
-                            lmmsize, cid, it, flags, reqp, cb_blocking);
+        sop_data->op_fid1 = rpid;
+
+        rc = md_intent_lock(lmv->tgts[mds].ltd_exp, sop_data, lmm,
+                            lmmsize, it, flags, reqp, cb_blocking,
+                            extra_lock_flags);
+
+        LASSERTF(rc != -ERESTART, "GETATTR: Got unhandled -ERESTART!\n");
         if (rc < 0)
-                RETURN(rc);
-       
+                GOTO(out_free_sop_data, rc);
+
         if (obj && rc > 0) {
-                /* this is splitted dir. In order to optimize things a
-                 * bit, we consider obj valid updating missing parts.
+                /*
+                 * This is split dir. In order to optimize things a bit, we
+                 * consider obj valid updating missing parts.
 
-                 * FIXME: do we need to return any lock here? It would
-                 * be fine if we don't. this means that nobody should
-                 * use UPDATE lock to notify about object * removal */
+                 * FIXME: do we need to return any lock here? It would be fine
+                 * if we don't. This means that nobody should use UPDATE lock to
+                 * notify about object * removal.
+                 */
                 CDEBUG(D_OTHER,
-                       "revalidate slaves for "DLID4", rc %d\n",
-                       OLID4(cid), rc);
-                
-                LASSERT(cid != 0);
-                rc = lmv_revalidate_slaves(exp, reqp, cid, it, rc,
-                                           cb_blocking);
-                RETURN(rc);
+                       "revalidate slaves for "DFID", rc %d\n",
+                       PFID(&op_data->op_fid2), rc);
+
+                LASSERT(fid_is_sane(&op_data->op_fid2));
+                rc = lmv_revalidate_slaves(exp, reqp, &op_data->op_fid2, it, rc,
+                                           cb_blocking, extra_lock_flags);
+                GOTO(out_free_sop_data, rc);
         }
 
         if (*reqp == NULL)
-                RETURN(rc);
-        /* okay, MDS has returned success. probably name has been
-         * resolved in remote inode */
-        rc = lmv_handle_remote_inode(exp, lmm, lmmsize, it,
-                                     flags, reqp, cb_blocking);
+                GOTO(out_free_sop_data, rc);
+
+        /*
+         * okay, MDS has returned success. Probably name has been resolved in
+         * remote inode.
+         */
+        rc = lmv_intent_remote(exp, lmm, lmmsize, it, flags,
+                               reqp, cb_blocking, extra_lock_flags);
         if (rc < 0)
-                RETURN(rc);
+                GOTO(out_free_sop_data, rc);
 
         /*
-         * nothing is found, do not access body->id1 as it is zero and thus
+         * Nothing is found, do not access body->fid1 as it is zero and thus
          * pointless.
          */
-        if (LUSTRE_IT(it)->it_disposition & DISP_LOOKUP_NEG)
-                RETURN(0);
-                
-        body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
+        if (it->d.lustre.it_disposition & DISP_LOOKUP_NEG)
+                GOTO(out_free_sop_data, rc = 0);
+
+        LASSERT(*reqp);
+        LASSERT((*reqp)->rq_repmsg);
+        body = lustre_msg_buf((*reqp)->rq_repmsg,
+                              DLM_REPLY_REC_OFF, sizeof(*body));
         LASSERT(body != NULL);
-        LASSERT((body->valid & OBD_MD_FID) != 0);
+        LASSERT(lustre_rep_swabbed(*reqp, DLM_REPLY_REC_OFF));
 
-        cid = &body->id1;
-        obj2 = lmv_grab_obj(obd, cid);
+        /* could not find object, FID is not present in response. */
+        if (!(body->valid & OBD_MD_FLID))
+                GOTO(out_free_sop_data, rc = 0);
 
-        if (!obj2 && (mea = lmv_splitted_dir_body(*reqp, 1))) {
-                /* wow! this is splitted dir, we'd like to handle it. */
-                body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
-                LASSERT(body != NULL);
+        obj2 = lmv_obj_grab(obd, &body->fid1);
+
+        if (!obj2 && (mea = lmv_get_mea(*reqp, DLM_REPLY_REC_OFF))) {
 
-                obj2 = lmv_create_obj(exp, &body->id1, mea);
+                /* FIXME remote capability! */
+                /* wow! this is split dir, we'd like to handle it. */
+                obj2 = lmv_obj_create(exp, &body->fid1, mea);
                 if (IS_ERR(obj2))
-                        RETURN(PTR_ERR(obj2));
+                        GOTO(out_free_sop_data, rc = (int)PTR_ERR(obj2));
         }
 
         if (obj2) {
-                /* this is splitted dir and we'd want to get attrs */
-                CDEBUG(D_OTHER, "attrs from slaves for "DLID4", rc %d\n",
-                       OLID4(cid), rc);
-                
-                rc = lmv_revalidate_slaves(exp, reqp, cid, it, 1,
-                                           cb_blocking);
-                lmv_put_obj(obj2);
+                /* this is split dir and we'd want to get attrs */
+                CDEBUG(D_OTHER, "attrs from slaves for "DFID", rc %d\n",
+                       PFID(&body->fid1), rc);
+
+                rc = lmv_revalidate_slaves(exp, reqp, &body->fid1, it, 1,
+                                           cb_blocking, extra_lock_flags);
+                lmv_obj_put(obj2);
         }
-        RETURN(rc);
-}
 
-void lmv_update_body_from_obj(struct mds_body *body, struct lmv_inode *obj)
-{
-        /* update size */
-        body->size += obj->size;
+        EXIT;
+out_free_sop_data:
+        OBD_FREE_PTR(sop_data);
+        return rc;
 }
 
+/* this is not used currently */
 int lmv_lookup_slaves(struct obd_export *exp, struct ptlrpc_request **reqp)
 {
         struct obd_device *obd = exp->exp_obd;
         struct lmv_obd *lmv = &obd->u.lmv;
-        struct mds_body *body = NULL;
         struct lustre_handle *lockh;
+        struct md_op_data *op_data;
         struct ldlm_lock *lock;
-        struct mds_body *body2;
+        struct mdt_body *body2;
+        struct mdt_body *body;
         struct lmv_obj *obj;
         int i, rc = 0;
         ENTRY;
@@ -372,49 +537,65 @@ int lmv_lookup_slaves(struct obd_export *exp, struct ptlrpc_request **reqp)
         LASSERT(reqp);
         LASSERT(*reqp);
 
-        /* master is locked. we'd like to take locks on slaves and update
+        /*
+         * Master is locked. we'd like to take locks on slaves and update
          * attributes to be returned from the slaves it's important that lookup
          * is called in two cases:
-         
-         *  - for first time (dcache has no such a resolving yet).
-         *  - ->d_revalidate() returned false.
-         
-         * last case possible only if all the objs (master and all slaves aren't
-         * valid */
-
-        body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
+
+         *  - for first time (dcache has no such a resolving yet).  -
+         *  ->d_revalidate() returned false.
+
+         * Last case possible only if all the objs (master and all slaves aren't
+         * valid.
+         */
+
+        OBD_ALLOC_PTR(op_data);
+        if (op_data == NULL)
+                RETURN(-ENOMEM);
+
+        body = lustre_msg_buf((*reqp)->rq_repmsg,
+                              DLM_REPLY_REC_OFF, sizeof(*body));
         LASSERT(body != NULL);
-        LASSERT((body->valid & OBD_MD_FID) != 0);
+        LASSERT(lustre_rep_swabbed(*reqp, DLM_REPLY_REC_OFF));
 
-        obj = lmv_grab_obj(obd, &body->id1);
+        LASSERT((body->valid & OBD_MD_FLID) != 0);
+        obj = lmv_obj_grab(obd, &body->fid1);
         LASSERT(obj != NULL);
 
-        CDEBUG(D_OTHER, "lookup slaves for "DLID4"\n", 
-               OLID4(&body->id1));
+        CDEBUG(D_OTHER, "lookup slaves for "DFID"\n",
+               PFID(&body->fid1));
 
-        lmv_lock_obj(obj);
-        
-        for (i = 0; i < obj->objcount; i++) {
-                struct lustre_id id = obj->objs[i].id;
+        lmv_obj_lock(obj);
+
+        for (i = 0; i < obj->lo_objcount; i++) {
+                struct lu_fid fid = obj->lo_inodes[i].li_fid;
                 struct ptlrpc_request *req = NULL;
+                struct obd_export *tgt_exp;
                 struct lookup_intent it;
 
-                if (id_equal_fid(&id, &obj->id))
+                if (lu_fid_eq(&fid, &obj->lo_fid))
                         /* skip master obj */
                         continue;
 
-                CDEBUG(D_OTHER, "lookup slave "DLID4"\n", OLID4(&id));
+                CDEBUG(D_OTHER, "lookup slave "DFID"\n", PFID(&fid));
 
                 /* is obj valid? */
                 memset(&it, 0, sizeof(it));
                 it.it_op = IT_GETATTR;
-                OBD_ALLOC(it.d.fs_data, sizeof(struct lustre_intent_data));
 
-                rc = md_intent_lock(lmv->tgts[id_group(&id)].ltd_exp, &id,
-                                    NULL, 0, NULL, 0, &id, &it, 0, &req,
-                                    lmv_dirobj_blocking_ast);
-                
-                lockh = (struct lustre_handle *)&LUSTRE_IT(&it)->it_lock_handle;
+                memset(op_data, 0, sizeof(*op_data));
+                op_data->op_fid1 = fid;
+                op_data->op_fid2 = fid;
+                op_data->op_bias = MDS_CROSS_REF;
+
+                tgt_exp = lmv_get_export(lmv, obj->lo_inodes[i].li_mds);
+                if (IS_ERR(tgt_exp))
+                        GOTO(cleanup, rc = PTR_ERR(tgt_exp));
+
+                rc = md_intent_lock(tgt_exp, op_data, NULL, 0, &it, 0,
+                                    &req, lmv_blocking_ast, 0);
+
+                lockh = (struct lustre_handle *)&it.d.lustre.it_lock_handle;
                 if (rc > 0 && req == NULL) {
                         /* nice, this slave is valid */
                         LASSERT(req == NULL);
@@ -423,259 +604,323 @@ int lmv_lookup_slaves(struct obd_export *exp, struct ptlrpc_request **reqp)
                 }
 
                 if (rc < 0) {
-                        OBD_FREE(it.d.fs_data, sizeof(struct lustre_intent_data));
                         /* error during lookup */
                         GOTO(cleanup, rc);
-                } 
+                }
                 lock = ldlm_handle2lock(lockh);
                 LASSERT(lock);
 
-                lock->l_ast_data = lmv_get_obj(obj);
+                lock->l_ast_data = lmv_obj_get(obj);
+
+                body2 = lustre_msg_buf(req->rq_repmsg,
+                                       DLM_REPLY_REC_OFF, sizeof(*body2));
+                LASSERT(body2 != NULL);
+                LASSERT(lustre_rep_swabbed(req, DLM_REPLY_REC_OFF));
 
-                body2 = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body2));
-                LASSERT(body2);
+                obj->lo_inodes[i].li_size = body2->size;
 
-                obj->objs[i].size = body2->size;
-                
                 CDEBUG(D_OTHER, "fresh: %lu\n",
-                       (unsigned long)obj->objs[i].size);
+                       (unsigned long)obj->lo_inodes[i].li_size);
 
                 LDLM_LOCK_PUT(lock);
 
                 if (req)
                         ptlrpc_req_finished(req);
 release_lock:
-                lmv_update_body_from_obj(body, obj->objs + i);
+                lmv_update_body(body, obj->lo_inodes + i);
 
-                if (LUSTRE_IT(&it)->it_lock_mode)
-                        ldlm_lock_decref(lockh, LUSTRE_IT(&it)->it_lock_mode);
-                OBD_FREE(it.d.fs_data, sizeof(struct lustre_intent_data));
+                if (it.d.lustre.it_lock_mode) {
+                        ldlm_lock_decref(lockh, it.d.lustre.it_lock_mode);
+                        it.d.lustre.it_lock_mode = 0;
+                }
         }
 
         EXIT;
 cleanup:
-        lmv_unlock_obj(obj);
-        lmv_put_obj(obj);
+        lmv_obj_unlock(obj);
+        lmv_obj_put(obj);
+        OBD_FREE_PTR(op_data);
         return rc;
 }
 
-int lmv_intent_lookup(struct obd_export *exp, struct lustre_id *pid,
-                      const char *name, int len, void *lmm, int lmmsize,
-                      struct lustre_id *cid, struct lookup_intent *it,
+int lmv_intent_lookup(struct obd_export *exp, struct md_op_data *op_data,
+                      void *lmm, int lmmsize, struct lookup_intent *it,
                       int flags, struct ptlrpc_request **reqp,
-                      ldlm_blocking_callback cb_blocking)
+                      ldlm_blocking_callback cb_blocking,
+                      int extra_lock_flags)
 {
         struct obd_device *obd = exp->exp_obd;
+        struct lu_fid rpid = op_data->op_fid1;
         struct lmv_obd *lmv = &obd->u.lmv;
-        struct mds_body *body = NULL;
-        struct lustre_id rpid = *pid;
+        struct md_op_data *sop_data;
+        struct lmv_stripe_md *mea;
+        struct mdt_body *body;
         struct lmv_obj *obj;
-        struct mea *mea;
-        int rc, mds, loop = 0;
+        int rc, loop = 0;
+        int mea_idx;
+        mdsno_t mds;
         ENTRY;
 
+        OBD_ALLOC_PTR(sop_data);
+        if (sop_data == NULL)
+                RETURN(-ENOMEM);
+
+        /* save op_data fro repeat case */
+        *sop_data = *op_data;
+
         /*
-         * IT_LOOKUP is intended to produce name -> id resolving (let's call
+         * IT_LOOKUP is intended to produce name -> fid resolving (let's call
          * this lookup below) or to confirm requested resolving is still valid
-         * (let's call this revalidation) cid != NULL specifies revalidation.
+         * (let's call this revalidation) fid_is_sane(&sop_data->op_fid2) specifies
+         * revalidation.
          */
-        if (cid) {
+        if (fid_is_sane(&op_data->op_fid2)) {
                 /*
-                 * this is revalidation: we have to check is LOOKUP lock still
-                 * valid for given id. Very important part is that we have to
+                 * This is revalidate: we have to check is LOOKUP lock still
+                 * valid for given fid. Very important part is that we have to
                  * choose right mds because namespace is per mds.
                  */
-                rpid = *pid;
-                obj = lmv_grab_obj(obd, pid);
+                rpid = op_data->op_fid1;
+                obj = lmv_obj_grab(obd, &rpid);
                 if (obj) {
-                        mds = raw_name2idx(obj->hashtype, obj->objcount,
-                                           (char *)name, len);
-                        rpid = obj->objs[mds].id;
-                        lmv_put_obj(obj);
+                        mea_idx = raw_name2idx(obj->lo_hashtype,
+                                               obj->lo_objcount,
+                                               (char *)op_data->op_name,
+                                               op_data->op_namelen);
+                        rpid = obj->lo_inodes[mea_idx].li_fid;
+                        mds = obj->lo_inodes[mea_idx].li_mds;
+                        sop_data->op_bias &= ~MDS_CHECK_SPLIT;
+                        lmv_obj_put(obj);
+                } else {
+                        rc = lmv_fld_lookup(lmv, &rpid, &mds);
+                        if (rc)
+                                GOTO(out_free_sop_data, rc);
+                        sop_data->op_bias |= MDS_CHECK_SPLIT;
                 }
-                mds = id_group(&rpid);
-
-                CDEBUG(D_OTHER, "revalidate lookup for "DLID4" to %d MDS\n",
-                       OLID4(cid), mds);
 
+                CDEBUG(D_OTHER, "revalidate lookup for "DFID" to #"LPU64" MDS\n",
+                       PFID(&op_data->op_fid2), mds);
         } else {
-                mds = id_group(pid);
 repeat:
-                LASSERT(++loop <= 2);
-                
-                /* this is lookup. during lookup we have to update all the
-                 * attributes, because returned values will be put in struct
-                 * inode */
+                ++loop;
+                LASSERT(loop <= 2);
 
-                obj = lmv_grab_obj(obd, pid);
+                /*
+                 * This is lookup. During lookup we have to update all the
+                 * attributes, because returned values will be put in struct
+                 * inode.
+                 */
+                obj = lmv_obj_grab(obd, &op_data->op_fid1);
                 if (obj) {
-                        if (len) {
-                                /* directory is already splitted. calculate mds */
-                                mds = raw_name2idx(obj->hashtype, obj->objcount, 
-                                                   (char *)name, len);
-                                rpid = obj->objs[mds].id;
-                                mds = id_group(&rpid);
+                        if (op_data->op_namelen) {
+                                /* directory is already split. calculate mds */
+                                mea_idx = raw_name2idx(obj->lo_hashtype,
+                                                       obj->lo_objcount,
+                                                       (char *)op_data->op_name,
+                                                       op_data->op_namelen);
+                                rpid = obj->lo_inodes[mea_idx].li_fid;
+                                mds = obj->lo_inodes[mea_idx].li_mds;
                         }
-                        lmv_put_obj(obj);
+                        sop_data->op_bias &= ~MDS_CHECK_SPLIT;
+                        lmv_obj_put(obj);
+                } else {
+                        rc = lmv_fld_lookup(lmv, &op_data->op_fid1, &mds);
+                        if (rc)
+                                GOTO(out_free_sop_data, rc);
+                        sop_data->op_bias |= MDS_CHECK_SPLIT;
                 }
+                fid_zero(&sop_data->op_fid2);
         }
-        rc = md_intent_lock(lmv->tgts[mds].ltd_exp, &rpid, name,
-                            len, lmm, lmmsize, cid, it, flags,
-                            reqp, cb_blocking);
-        if (rc > 0) {
-                LASSERT(cid != 0);
-                RETURN(rc);
-        }
+
+        sop_data->op_bias &= ~MDS_CROSS_REF;
+        sop_data->op_fid1 = rpid;
+
+        rc = md_intent_lock(lmv->tgts[mds].ltd_exp, sop_data, lmm, lmmsize,
+                            it, flags, reqp, cb_blocking, extra_lock_flags);
         if (rc > 0) {
-                /* very interesting. it seems object is still valid but for some
-                 * reason llite calls lookup, not revalidate */
-                CDEBUG(D_OTHER, "lookup for "DLID4" and data should be uptodate\n",
-                      OLID4(&rpid));
+                LASSERT(fid_is_sane(&op_data->op_fid2));
+                /*
+                 * Very interesting. it seems object is still valid but for some
+                 * reason llite calls lookup, not revalidate.
+                 */
+                CDEBUG(D_OTHER, "lookup for "DFID" and data should be uptodate\n",
+                       PFID(&rpid));
                 LASSERT(*reqp == NULL);
-                RETURN(rc);
+                GOTO(out_free_sop_data, rc);
         }
 
         if (rc == 0 && *reqp == NULL) {
                 /* once again, we're asked for lookup, not revalidate */
-                CDEBUG(D_OTHER, "lookup for "DLID4" and data should be uptodate\n",
-                      OLID4(&rpid));
-                RETURN(rc);
+                CDEBUG(D_OTHER, "lookup for "DFID" and data should be uptodate\n",
+                       PFID(&rpid));
+                GOTO(out_free_sop_data, rc);
         }
-       
+
         if (rc == -ERESTART) {
-                /* directory got splitted since last update. this shouldn't be
-                 * becasue splitting causes lock revocation, so revalidate had
-                 * to fail and lookup on dir had to return mea */
+                LASSERT(*reqp != NULL);
+                DEBUG_REQ(D_WARNING|D_RPCTRACE, *reqp,
+                          "Got -ERESTART during lookup!\n");
+                ptlrpc_req_finished(*reqp);
+                *reqp = NULL;
+                it->d.lustre.it_data = 0;
+                /*
+                 * Directory got split since last update. This shouldn't be
+                 * because splitting causes lock revocation, so revalidate had
+                 * to fail and lookup on dir had to return mea.
+                 */
                 CWARN("we haven't knew about directory splitting!\n");
                 LASSERT(obj == NULL);
 
-                obj = lmv_create_obj(exp, &rpid, NULL);
+                obj = lmv_obj_create(exp, &rpid, NULL);
                 if (IS_ERR(obj))
-                        RETURN(PTR_ERR(obj));
-                lmv_put_obj(obj);
+                        GOTO(out_free_sop_data, rc = PTR_ERR(obj));
+                lmv_obj_put(obj);
                 goto repeat;
         }
 
         if (rc < 0)
-                RETURN(rc);
+                GOTO(out_free_sop_data, rc);
 
-        /* okay, MDS has returned success. Probably name has been resolved in
-         * remote inode. */
-        rc = lmv_handle_remote_inode(exp, lmm, lmmsize, it,
-                                     flags, reqp, cb_blocking);
+        /*
+         * Okay, MDS has returned success. Probably name has been resolved in
+         * remote inode.
+         */
+        rc = lmv_intent_remote(exp, lmm, lmmsize, it, flags, reqp,
+                               cb_blocking, extra_lock_flags);
 
-        if (rc == 0 && (mea = lmv_splitted_dir_body(*reqp, 1))) {
-                /* wow! this is splitted dir, we'd like to handle it */
-                body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
+        if (rc == 0 && (mea = lmv_get_mea(*reqp, DLM_REPLY_REC_OFF))) {
+                /* Wow! This is split dir, we'd like to handle it. */
+                body = lustre_msg_buf((*reqp)->rq_repmsg,
+                                      DLM_REPLY_REC_OFF, sizeof(*body));
                 LASSERT(body != NULL);
-                LASSERT((body->valid & OBD_MD_FID) != 0);
-                
-                obj = lmv_grab_obj(obd, &body->id1);
+                LASSERT(lustre_rep_swabbed(*reqp, DLM_REPLY_REC_OFF));
+                LASSERT((body->valid & OBD_MD_FLID) != 0);
+
+                obj = lmv_obj_grab(obd, &body->fid1);
                 if (!obj) {
-                        obj = lmv_create_obj(exp, &body->id1, mea);
+                        obj = lmv_obj_create(exp, &body->fid1, mea);
                         if (IS_ERR(obj))
-                                RETURN(PTR_ERR(obj));
+                                GOTO(out_free_sop_data, rc = (int)PTR_ERR(obj));
                 }
-                lmv_put_obj(obj);
+                lmv_obj_put(obj);
         }
 
-        RETURN(rc);
+        EXIT;
+out_free_sop_data:
+        OBD_FREE_PTR(sop_data);
+        return rc;
 }
 
-int lmv_intent_lock(struct obd_export *exp, struct lustre_id *pid,
-                    const char *name, int len, void *lmm, int lmmsize,
-                    struct lustre_id *cid, struct lookup_intent *it,
+int lmv_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
+                    void *lmm, int lmmsize, struct lookup_intent *it,
                     int flags, struct ptlrpc_request **reqp,
-                    ldlm_blocking_callback cb_blocking)
+                    ldlm_blocking_callback cb_blocking,
+                    int extra_lock_flags)
 {
         struct obd_device *obd = exp->exp_obd;
-        int rc = 0;
+        int rc;
         ENTRY;
 
-        LASSERT(it);
-        LASSERT(pid);
+        LASSERT(it != NULL);
+        LASSERT(fid_is_sane(&op_data->op_fid1));
 
-        CDEBUG(D_OTHER, "INTENT LOCK '%s' for '%*s' on %lu/%lu -> %lu\n",
-               LL_IT2STR(it), len, name, (unsigned long)id_ino(pid),
-               (unsigned long)id_gen(pid), (unsigned long)id_group(pid));
+        CDEBUG(D_OTHER, "INTENT LOCK '%s' for '%*s' on "DFID"\n",
+               LL_IT2STR(it), op_data->op_namelen, op_data->op_name,
+               PFID(&op_data->op_fid1));
 
         rc = lmv_check_connect(obd);
         if (rc)
                 RETURN(rc);
 
-        if (it->it_op == IT_LOOKUP)
-                rc = lmv_intent_lookup(exp, pid, name, len, lmm,
-                                       lmmsize, cid, it, flags, reqp,
-                                       cb_blocking);
+        if (it->it_op & IT_LOOKUP)
+                rc = lmv_intent_lookup(exp, op_data, lmm, lmmsize, it,
+                                       flags, reqp, cb_blocking,
+                                       extra_lock_flags);
         else if (it->it_op & IT_OPEN)
-                rc = lmv_intent_open(exp, pid, name, len, lmm,
-                                     lmmsize, cid, it, flags, reqp,
-                                     cb_blocking);
-        else if (it->it_op == IT_GETATTR || it->it_op == IT_CHDIR)
-                rc = lmv_intent_getattr(exp, pid, name, len, lmm,
-                                        lmmsize, cid, it, flags, reqp,
-                                        cb_blocking);
+                rc = lmv_intent_open(exp, op_data, lmm, lmmsize, it,
+                                     flags, reqp, cb_blocking,
+                                     extra_lock_flags);
+        else if (it->it_op & IT_GETATTR)
+                rc = lmv_intent_getattr(exp, op_data,lmm, lmmsize, it,
+                                        flags, reqp, cb_blocking,
+                                        extra_lock_flags);
         else
                 LBUG();
         RETURN(rc);
 }
 
 int lmv_revalidate_slaves(struct obd_export *exp, struct ptlrpc_request **reqp,
-                          struct lustre_id *mid, struct lookup_intent *oit,
-                          int master_valid, ldlm_blocking_callback cb_blocking)
+                          const struct lu_fid *mid, struct lookup_intent *oit,
+                          int master_valid, ldlm_blocking_callback cb_blocking,
+                          int extra_lock_flags)
 {
         struct obd_device *obd = exp->exp_obd;
         struct ptlrpc_request *mreq = *reqp;
         struct lmv_obd *lmv = &obd->u.lmv;
         struct lustre_handle master_lockh;
+        struct obd_export *tgt_exp;
+        struct md_op_data *op_data;
         struct ldlm_lock *lock;
         unsigned long size = 0;
-        struct mds_body *body;
+        struct mdt_body *body;
         struct lmv_obj *obj;
         int master_lock_mode;
         int i, rc = 0;
         ENTRY;
 
-        /* we have to loop over the subobjects, check validity and update them
+        OBD_ALLOC_PTR(op_data);
+        if (op_data == NULL)
+                RETURN(-ENOMEM);
+
+        /*
+         * We have to loop over the subobjects, check validity and update them
          * from MDSs if needed. it's very useful that we need not to update all
          * the fields. say, common fields (that are equal on all the subojects
          * need not to be update, another fields (i_size, for example) are
-         * cached all the time */
-        obj = lmv_grab_obj(obd, mid);
+         * cached all the time.
+         */
+        obj = lmv_obj_grab(obd, mid);
         LASSERT(obj != NULL);
 
         master_lock_mode = 0;
 
-        lmv_lock_obj(obj);
-        
-        for (i = 0; i < obj->objcount; i++) {
-                struct lustre_id id = obj->objs[i].id;
+        lmv_obj_lock(obj);
+
+        for (i = 0; i < obj->lo_objcount; i++) {
+                struct lu_fid fid = obj->lo_inodes[i].li_fid;
                 struct lustre_handle *lockh = NULL;
                 struct ptlrpc_request *req = NULL;
                 ldlm_blocking_callback cb;
                 struct lookup_intent it;
                 int master = 0;
 
-                CDEBUG(D_OTHER, "revalidate subobj "DLID4"\n",
-                       OLID4(&id));
+                CDEBUG(D_OTHER, "revalidate subobj "DFID"\n",
+                       PFID(&fid));
 
+                memset(op_data, 0, sizeof(*op_data));
                 memset(&it, 0, sizeof(it));
                 it.it_op = IT_GETATTR;
 
-                cb = lmv_dirobj_blocking_ast;
+                cb = lmv_blocking_ast;
 
-                OBD_ALLOC(it.d.fs_data, sizeof(struct lustre_intent_data));
-                if (id_equal_fid(&id, &obj->id)) {
+                if (lu_fid_eq(&fid, &obj->lo_fid)) {
                         if (master_valid) {
-                                /* lmv_intent_getattr() already checked
-                                 * validness and took the lock */
+                                /*
+                                 * lmv_intent_getattr() already checked
+                                 * validness and took the lock.
+                                 */
                                 if (mreq) {
-                                        /* it even got the reply refresh attrs
-                                         * from that reply */
+                                        /*
+                                         * It even got the reply refresh attrs
+                                         * from that reply.
+                                         */
                                         body = lustre_msg_buf(mreq->rq_repmsg,
-                                                              1, sizeof(*body));
+                                                              DLM_REPLY_REC_OFF,
+                                                              sizeof(*body));
                                         LASSERT(body != NULL);
-                                        goto update; 
+                                        LASSERT(lustre_rep_swabbed(
+                                                      mreq, DLM_REPLY_REC_OFF));
+                                        goto update;
                                 }
                                 /* take already cached attrs into account */
                                 CDEBUG(D_OTHER,
@@ -686,104 +931,117 @@ int lmv_revalidate_slaves(struct obd_export *exp, struct ptlrpc_request **reqp,
                         cb = cb_blocking;
                 }
 
-               
-                /* is obj valid? */
-                rc = md_intent_lock(lmv->tgts[id_group(&id)].ltd_exp,
-                                    &id, NULL, 0, NULL, 0, &id, &it, 0, 
-                                    &req, cb);
-                lockh = (struct lustre_handle *) &LUSTRE_IT(&it)->it_lock_handle;
+                op_data->op_fid1 = fid;
+                op_data->op_fid2 = fid;
+                op_data->op_bias = MDS_CROSS_REF;
+
+                /* Is obj valid? */
+                tgt_exp = lmv_get_export(lmv, obj->lo_inodes[i].li_mds);
+                if (IS_ERR(tgt_exp))
+                        GOTO(cleanup, rc = PTR_ERR(tgt_exp));
+
+                rc = md_intent_lock(tgt_exp, op_data, NULL, 0, &it, 0, &req, cb,
+                                    extra_lock_flags);
+
+                lockh = (struct lustre_handle *)&it.d.lustre.it_lock_handle;
                 if (rc > 0 && req == NULL) {
-                        /* nice, this slave is valid */
+                        /* Nice, this slave is valid */
                         LASSERT(req == NULL);
                         CDEBUG(D_OTHER, "cached\n");
                         goto release_lock;
                 }
 
-                if (rc < 0) {
-                        OBD_FREE(it.d.fs_data, sizeof(struct lustre_intent_data));
-                        /* error during revalidation */
+                if (rc < 0)
                         GOTO(cleanup, rc);
-                }
+
                 if (master) {
                         LASSERT(master_valid == 0);
-                        /* save lock on master to be returned to the caller */
+                        /* Save lock on master to be returned to the caller. */
                         CDEBUG(D_OTHER, "no lock on master yet\n");
                         memcpy(&master_lockh, lockh, sizeof(master_lockh));
-                        master_lock_mode = LUSTRE_IT(&it)->it_lock_mode;
-                        LUSTRE_IT(&it)->it_lock_mode = 0;
+                        master_lock_mode = it.d.lustre.it_lock_mode;
+                        it.d.lustre.it_lock_mode = 0;
                 } else {
-                        /* this is slave. we want to control it */
+                        /* This is slave. We want to control it. */
                         lock = ldlm_handle2lock(lockh);
-                        LASSERT(lock);
-                        lock->l_ast_data = lmv_get_obj(obj);
+                        LASSERT(lock != NULL);
+                        lock->l_ast_data = lmv_obj_get(obj);
                         LDLM_LOCK_PUT(lock);
                 }
 
                 if (*reqp == NULL) {
-                        /* this is first reply, we'll use it to return updated
-                         * data back to the caller */
+                        /*
+                         * This is first reply, we'll use it to return updated
+                         * data back to the caller.
+                         */
                         LASSERT(req);
                         ptlrpc_request_addref(req);
                         *reqp = req;
-
                 }
 
-                body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
-                LASSERT(body);
-                
+                body = lustre_msg_buf(req->rq_repmsg,
+                                      DLM_REPLY_REC_OFF, sizeof(*body));
+                LASSERT(body != NULL);
+                LASSERT(lustre_rep_swabbed(req, DLM_REPLY_REC_OFF));
+
 update:
-                obj->objs[i].size = body->size;
-                
+                obj->lo_inodes[i].li_size = body->size;
+
                 CDEBUG(D_OTHER, "fresh: %lu\n",
-                       (unsigned long)obj->objs[i].size);
-                
+                       (unsigned long)obj->lo_inodes[i].li_size);
+
                 if (req)
                         ptlrpc_req_finished(req);
 release_lock:
-                size += obj->objs[i].size;
+                size += obj->lo_inodes[i].li_size;
 
-                if (LUSTRE_IT(&it)->it_lock_mode)
-                        ldlm_lock_decref(lockh, LUSTRE_IT(&it)->it_lock_mode);
-                OBD_FREE(it.d.fs_data, sizeof(struct lustre_intent_data));
+                if (it.d.lustre.it_lock_mode) {
+                        ldlm_lock_decref(lockh, it.d.lustre.it_lock_mode);
+                        it.d.lustre.it_lock_mode = 0;
+                }
         }
 
         if (*reqp) {
-                /* some attrs got refreshed, we have reply and it's time to put
-                 * fresh attrs to it */
+                /*
+                 * Some attrs got refreshed, we have reply and it's time to put
+                 * fresh attrs to it.
+                 */
                 CDEBUG(D_OTHER, "return refreshed attrs: size = %lu\n",
                        (unsigned long)size);
-                
-                body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
-                LASSERT(body);
 
-                /* FIXME: what about other attributes? */
+                body = lustre_msg_buf((*reqp)->rq_repmsg,
+                                      DLM_REPLY_REC_OFF, sizeof(*body));
+                LASSERT(body != NULL);
+                LASSERT(lustre_rep_swabbed(*reqp, DLM_REPLY_REC_OFF));
+
                 body->size = size;
-                
+
                 if (mreq == NULL) {
-                        /* very important to maintain id_group(lli->lli_id) the
-                         * same because of revalidation. mreq == NULL means that
-                         * caller has no reply and the only attr we can return
-                         * is size */
+                        /*
+                         * Very important to maintain mds num the same because
+                         * of revalidation. mreq == NULL means that caller has
+                         * no reply and the only attr we can return is size.
+                         */
                         body->valid = OBD_MD_FLSIZE;
-//                        body->mds = id_group(&obj->id);
                 }
                 if (master_valid == 0) {
-                        memcpy(&LUSTRE_IT(oit)->it_lock_handle,
+                        memcpy(&oit->d.lustre.it_lock_handle,
                                &master_lockh, sizeof(master_lockh));
-                        LUSTRE_IT(oit)->it_lock_mode = master_lock_mode;
+                        oit->d.lustre.it_lock_mode = master_lock_mode;
                 }
                 rc = 0;
         } else {
-                /* it seems all the attrs are fresh and we did no request */
+                /* It seems all the attrs are fresh and we did no request */
                 CDEBUG(D_OTHER, "all the attrs were fresh\n");
                 if (master_valid == 0)
-                        LUSTRE_IT(oit)->it_lock_mode = master_lock_mode;
+                        oit->d.lustre.it_lock_mode = master_lock_mode;
                 rc = 1;
         }
 
         EXIT;
 cleanup:
-        lmv_unlock_obj(obj);
-        lmv_put_obj(obj);
+        OBD_FREE_PTR(op_data);
+        lmv_obj_unlock(obj);
+        lmv_obj_put(obj);
         return rc;
 }