Whamcloud - gitweb
- CROW-related fixes from b_hd_mdref
[fs/lustre-release.git] / lustre / lov / lov_obd.c
index fb34c8b..3186999 100644 (file)
 
 #include "lov_internal.h"
 
-static int lov_stripe_offset(struct lov_stripe_md *lsm, obd_off lov_off,
-                             int stripeno, obd_off *obd_off);
-
-struct lov_lock_handles {
-        struct portals_handle llh_handle;
-        atomic_t llh_refcount;
-        int llh_stripe_count;
-        struct lustre_handle llh_handles[0];
-};
-
-static void lov_llh_addref(void *llhp)
-{
-        struct lov_lock_handles *llh = llhp;
-
-        atomic_inc(&llh->llh_refcount);
-        CDEBUG(D_INFO, "GETting llh %p : new refcount %d\n", llh,
-               atomic_read(&llh->llh_refcount));
-}
-
-static struct lov_lock_handles *lov_llh_new(struct lov_stripe_md *lsm)
-{
-        struct lov_lock_handles *llh;
-
-        OBD_ALLOC(llh, sizeof *llh +
-                  sizeof(*llh->llh_handles) * lsm->lsm_stripe_count);
-        if (llh == NULL) {
-                CERROR("out of memory\n");
-                return NULL;
-        }
-        atomic_set(&llh->llh_refcount, 2);
-        llh->llh_stripe_count = lsm->lsm_stripe_count;
-        INIT_LIST_HEAD(&llh->llh_handle.h_link);
-        class_handle_hash(&llh->llh_handle, lov_llh_addref);
-        return llh;
-}
-
-static struct lov_lock_handles *lov_handle2llh(struct lustre_handle *handle)
-{
-        ENTRY;
-        LASSERT(handle != NULL);
-        RETURN(class_handle2object(handle->cookie));
-}
-
-static void lov_llh_put(struct lov_lock_handles *llh)
-{
-        CDEBUG(D_INFO, "PUTting llh %p : new refcount %d\n", llh,
-               atomic_read(&llh->llh_refcount) - 1);
-        LASSERT(atomic_read(&llh->llh_refcount) > 0 &&
-                atomic_read(&llh->llh_refcount) < 0x5a5a);
-        if (atomic_dec_and_test(&llh->llh_refcount)) {
-                LASSERT(list_empty(&llh->llh_handle.h_link));
-                OBD_FREE(llh, sizeof *llh +
-                         sizeof(*llh->llh_handles) * llh->llh_stripe_count);
-        }
-}
-
-static void lov_llh_destroy(struct lov_lock_handles *llh)
-{
-        class_handle_unhash(&llh->llh_handle);
-        lov_llh_put(llh);
-}
-
 /* obd methods */
 #define MAX_STRING_SIZE 128
 static int lov_connect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
-                           int activate, unsigned long connect_flags)
+                           int activate, struct obd_connect_data *conn_data,
+                           unsigned long connect_flags)
 {
-        struct lov_obd *lov = &obd->u.lov;
-        struct obd_uuid *tgt_uuid = &tgt->uuid;
-        struct obd_device *tgt_obd;
         struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
-        struct lustre_handle conn = {0, };
+        struct obd_uuid *tgt_uuid = &tgt->uuid;
+
+#ifdef __KERNEL__
         struct proc_dir_entry *lov_proc_dir;
+#endif
+        struct lov_obd *lov = &obd->u.lov;
+        struct lustre_handle conn = {0, };
+        struct obd_device *tgt_obd;
         int rc;
         ENTRY;
 
@@ -157,7 +99,8 @@ static int lov_connect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
                 RETURN(0);
         }
 
-        rc = obd_connect(&conn, tgt_obd, &lov_osc_uuid, connect_flags);
+        rc = obd_connect(&conn, tgt_obd, &lov_osc_uuid, conn_data,
+                         connect_flags);
         if (rc) {
                 CERROR("Target %s connect error %d\n", tgt_uuid->uuid, rc);
                 RETURN(rc);
@@ -176,6 +119,7 @@ static int lov_connect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
         tgt->active = 1;
         lov->desc.ld_active_tgt_count++;
 
+#ifdef __KERNEL__
         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
         if (lov_proc_dir) {
                 struct obd_device *osc_obd = class_conn2obd(&conn);
@@ -200,17 +144,21 @@ static int lov_connect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
                         lov_proc_dir = NULL;
                 }
         }
+#endif
 
         RETURN(0);
 }
 
 static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
-                       struct obd_uuid *cluuid, unsigned long connect_flags)
+                       struct obd_uuid *cluuid, struct obd_connect_data *data,
+                       unsigned long flags)
 {
+#ifdef __KERNEL__
+        struct proc_dir_entry *lov_proc_dir;
+#endif
         struct lov_obd *lov = &obd->u.lov;
         struct lov_tgt_desc *tgt;
         struct obd_export *exp;
-        struct proc_dir_entry *lov_proc_dir;
         int rc, rc2, i;
         ENTRY;
 
@@ -228,6 +176,7 @@ static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
                 RETURN(0);
         }
 
+#ifdef __KERNEL__
         lov_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
                                         NULL, NULL);
         if (IS_ERR(lov_proc_dir)) {
@@ -235,13 +184,14 @@ static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
                        obd->obd_type->typ_name, obd->obd_name);
                 lov_proc_dir = NULL;
         }
+#endif
 
         /* connect_flags is the MDS number, save for use in lov_add_obd */
-        lov->lov_connect_flags = connect_flags;
+        lov->lov_connect_flags = flags;
         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
                 if (obd_uuid_empty(&tgt->uuid))
                         continue;
-                rc = lov_connect_obd(obd, tgt, 0, connect_flags);
+                rc = lov_connect_obd(obd, tgt, 0, data, flags);
                 if (rc)
                         GOTO(out_disc, rc);
         }
@@ -250,8 +200,10 @@ static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
         RETURN (0);
 
  out_disc:
+#ifdef __KERNEL__
         if (lov_proc_dir)
                 lprocfs_remove(lov_proc_dir);
+#endif
 
         while (i-- > 0) {
                 struct obd_uuid uuid;
@@ -269,15 +221,19 @@ static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
         RETURN (rc);
 }
 
-static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
-                              int flags)
+static int lov_disconnect_obd(struct obd_device *obd, 
+                             struct lov_tgt_desc *tgt,
+                              unsigned long flags)
 {
+#ifdef __KERNEL__
         struct proc_dir_entry *lov_proc_dir;
+#endif
         struct obd_device *osc_obd = class_exp2obd(tgt->ltd_exp);
         struct lov_obd *lov = &obd->u.lov;
         int rc;
         ENTRY;
 
+#ifdef __KERNEL__
         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
         if (lov_proc_dir) {
                 struct proc_dir_entry *osc_symlink;
@@ -291,7 +247,7 @@ static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
                                osc_obd->obd_name);
                 }
         }
-
+#endif
         if (obd->obd_no_recov) {
                 /* Pass it on to our clients.
                  * XXX This should be an argument to disconnect,
@@ -302,7 +258,6 @@ static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
         }
 
         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
-
         rc = obd_disconnect(tgt->ltd_exp, flags);
         if (rc) {
                 if (tgt->active) {
@@ -316,17 +271,18 @@ static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
                 tgt->active = 0;
                 lov->desc.ld_active_tgt_count--;
         }
-
         tgt->ltd_exp = NULL;
         RETURN(0);
 }
 
-static int lov_disconnect(struct obd_export *exp, int flags)
+static int lov_disconnect(struct obd_export *exp, unsigned long flags)
 {
         struct obd_device *obd = class_exp2obd(exp);
+#ifdef __KERNEL__
+        struct proc_dir_entry *lov_proc_dir;
+#endif
         struct lov_obd *lov = &obd->u.lov;
         struct lov_tgt_desc *tgt;
-        struct proc_dir_entry *lov_proc_dir;
         int rc, i;
         ENTRY;
 
@@ -342,6 +298,8 @@ static int lov_disconnect(struct obd_export *exp, int flags)
                 if (tgt->ltd_exp)
                         lov_disconnect_obd(obd, tgt, flags);
         }
+
+#ifdef __KERNEL__
         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
         if (lov_proc_dir) {
                 lprocfs_remove(lov_proc_dir);
@@ -349,8 +307,8 @@ static int lov_disconnect(struct obd_export *exp, int flags)
                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing.",
                        obd->obd_type->typ_name, obd->obd_name);
         }
-
-
+#endif
+        
  out_local:
         rc = class_disconnect(exp, 0);
         RETURN(rc);
@@ -374,9 +332,13 @@ static int lov_set_osc_active(struct lov_obd *lov, struct obd_uuid *uuid,
 
         spin_lock(&lov->lov_lock);
         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
+                if (tgt->ltd_exp == NULL)
+                        continue;
+
                 CDEBUG(D_INFO, "lov 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;
         }
 
@@ -408,8 +370,8 @@ static int lov_set_osc_active(struct lov_obd *lov, struct obd_uuid *uuid,
 static int lov_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_OSC_NAME)) {
@@ -468,21 +430,21 @@ int lov_detach(struct obd_device *dev)
 
 static int lov_setup(struct obd_device *obd, obd_count len, void *buf)
 {
+        struct lov_obd *lov = &obd->u.lov;
         struct lustre_cfg *lcfg = buf;
         struct lov_desc *desc;
-        struct lov_obd *lov = &obd->u.lov;
         int count;
         ENTRY;
-                                                                                                                                                                                                     
-        if (lcfg->lcfg_inllen1 < 1) {
+
+        if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
                 CERROR("LOV setup requires a descriptor\n");
                 RETURN(-EINVAL);
         }
-                                                                                                                                                                                                     
-        desc = (struct lov_desc *)lcfg->lcfg_inlbuf1;
-        if (sizeof(*desc) > lcfg->lcfg_inllen1) {
+
+        desc = (struct lov_desc *)lustre_cfg_string(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);
         }
  
@@ -509,13 +471,13 @@ static int lov_setup(struct obd_device *obd, obd_count len, void *buf)
                        lov->bufsize);
                 RETURN(-EINVAL);
         }
-                                                                                                                                                                                                     
+
         desc->ld_tgt_count = 0;
         desc->ld_active_tgt_count = 0;
         lov->desc = *desc;
         spin_lock_init(&lov->lov_lock);
         sema_init(&lov->lov_llog_sem, 1);
-                                                                                                                                                                                                     
+
         RETURN(0);
 }
 
@@ -578,7 +540,7 @@ lov_add_obd(struct obd_device *obd, struct obd_uuid *uuidp, int index, int gen)
                         osc_obd->obd_no_recov = 0;
         }
 
-        rc = lov_connect_obd(obd, tgt, 1, lov->lov_connect_flags);
+        rc = lov_connect_obd(obd, tgt, 1, NULL, lov->lov_connect_flags);
         if (rc)
                 GOTO(out, rc);
 
@@ -620,7 +582,7 @@ lov_del_obd(struct obd_device *obd, struct obd_uuid *uuidp, int index, int gen)
                 RETURN(-EINVAL);
         }
 
-        if (strncmp(uuidp->uuid, tgt->uuid.uuid, sizeof uuidp->uuid) != 0) {
+        if (!obd_uuid_equals(uuidp, &tgt->uuid)) {
                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
                        tgt->uuid.uuid, index, uuidp->uuid);
                 RETURN(-EINVAL);
@@ -665,14 +627,14 @@ static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
         switch(cmd = lcfg->lcfg_command) {
         case LCFG_LOV_ADD_OBD:
         case LCFG_LOV_DEL_OBD: {
-                if (lcfg->lcfg_inllen1 > sizeof(obd_uuid.uuid))
+                if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
                         GOTO(out, rc = -EINVAL);
 
-                obd_str2uuid(&obd_uuid, lcfg->lcfg_inlbuf1);
+                obd_str2uuid(&obd_uuid, lustre_cfg_string(lcfg, 1));
 
-                if (sscanf(lcfg->lcfg_inlbuf2, "%d", &index) != 1)
+                if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
                         GOTO(out, rc = -EINVAL);
-                if (sscanf(lcfg->lcfg_inlbuf3, "%d", &gen) != 1)
+                if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
                         GOTO(out, rc = -EINVAL);
                 if (cmd == LCFG_LOV_ADD_OBD)
                         rc = lov_add_obd(obd, &obd_uuid, index, gen);
@@ -690,64 +652,12 @@ out:
         RETURN(rc);
 }
 
-/* compute object size given "stripeno" and the ost size */
-static obd_size lov_stripe_size(struct lov_stripe_md *lsm, obd_size ost_size,
-                                int stripeno)
-{
-        unsigned long ssize  = lsm->lsm_stripe_size;
-        unsigned long swidth = ssize * lsm->lsm_stripe_count;
-        unsigned long stripe_size;
-        obd_size lov_size;
-
-        if (ost_size == 0)
-                return 0;
-
-        /* do_div(a, b) returns a % b, and a = a / b */
-        stripe_size = do_div(ost_size, ssize);
-        if (stripe_size)
-                lov_size = ost_size * swidth + stripeno * ssize + stripe_size;
-        else
-                lov_size = (ost_size - 1) * swidth + (stripeno + 1) * ssize;
-
-        return lov_size;
-}
-
-static void lov_merge_attrs(struct obdo *tgt, struct obdo *src, obd_flag valid,
-                            struct lov_stripe_md *lsm, int stripeno, int *set)
-{
-        valid &= src->o_valid;
-
-        if (*set) {
-                if (valid & OBD_MD_FLSIZE) {
-                        /* this handles sparse files properly */
-                        obd_size lov_size;
-
-                        lov_size = lov_stripe_size(lsm, src->o_size, stripeno);
-                        if (lov_size > tgt->o_size)
-                                tgt->o_size = lov_size;
-                }
-                if (valid & OBD_MD_FLBLOCKS)
-                        tgt->o_blocks += src->o_blocks;
-                if (valid & OBD_MD_FLBLKSZ)
-                        tgt->o_blksize += src->o_blksize;
-                if (valid & OBD_MD_FLCTIME && tgt->o_ctime < src->o_ctime)
-                        tgt->o_ctime = src->o_ctime;
-                if (valid & OBD_MD_FLMTIME && tgt->o_mtime < src->o_mtime)
-                        tgt->o_mtime = src->o_mtime;
-        } else {
-                memcpy(tgt, src, sizeof(*tgt));
-                tgt->o_id = lsm->lsm_object_id;
-                if (valid & OBD_MD_FLSIZE)
-                        tgt->o_size = lov_stripe_size(lsm,src->o_size,stripeno);
-                *set = 1;
-        }
-}
-
 #ifndef log2
 #define log2(n) ffz(~(n))
 #endif
 
-static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
+static int lov_clear_orphans(struct obd_export *export,
+                             struct obdo *src_oa,
                              struct lov_stripe_md **ea,
                              struct obd_trans_info *oti)
 {
@@ -773,12 +683,14 @@ static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
         }
 
         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
+                int err;
                 struct lov_stripe_md obj_md;
                 struct lov_stripe_md *obj_mdp = &obj_md;
-                int err;
 
-                /* if called for a specific target, we don't
-                   care if it is not active. */
+                /*
+                 * if called for a specific target, we don't care if it is not
+                 * active.
+                 */
                 if (lov->tgts[i].active == 0 && ost_uuid == NULL) {
                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
                         continue;
@@ -787,15 +699,25 @@ static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &lov->tgts[i].uuid))
                         continue;
 
+                /* 
+                 * setting up objid OSS objects should be destroyed starting
+                 * from it.
+                 */
                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
+                tmp_oa->o_valid |= OBD_MD_FLID;
+                tmp_oa->o_id = oti->oti_objid[i];
 
                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
-                err = obd_create(lov->tgts[i].ltd_exp, tmp_oa, &obj_mdp, oti);
-                if (err)
-                        /* This export will be disabled until it is recovered,
-                           and then orphan recovery will be completed. */
+                err = obd_create(lov->tgts[i].ltd_exp, tmp_oa, NULL, 0,
+                                 &obj_mdp, oti);
+                if (err) {
+                        /*
+                         * this export will be disabled until it is recovered,
+                         * and then orphan recovery will be completed.
+                         */
                         CERROR("error in orphan recovery on OST idx %d/%d: "
                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
+                }
 
                 if (ost_uuid)
                         break;
@@ -804,23 +726,21 @@ static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
         RETURN(rc);
 }
 
-#define LOV_CREATE_RESEED_INTERVAL 1000
-
 /* the LOV expects oa->o_id to be set to the LOV object id */
-static int lov_create(struct obd_export *exp, struct obdo *src_oa,
-                      struct lov_stripe_md **ea, struct obd_trans_info *oti)
+static int
+lov_create(struct obd_export *exp, struct obdo *src_oa,
+           void *acl, int acl_size, struct lov_stripe_md **ea,
+           struct obd_trans_info *oti)
 {
-        static int ost_start_idx, ost_start_count;
+        struct lov_request_set *set = NULL;
+        struct list_head *pos;
         struct lov_obd *lov;
-        struct lov_stripe_md *lsm;
-        struct lov_oinfo *loi = NULL;
-        struct obdo *tmp_oa, *ret_oa;
-        struct llog_cookie *cookies = NULL;
-        unsigned ost_count, ost_idx;
-        int set = 0, obj_alloc = 0, cookie_sent = 0, rc = 0, i;
+        int rc = 0;
         ENTRY;
 
         LASSERT(ea != NULL);
+        if (exp == NULL)
+                RETURN(-EINVAL);
 
         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
             src_oa->o_flags == OBD_FL_DELORPHAN) {
@@ -828,324 +748,28 @@ static int lov_create(struct obd_export *exp, struct obdo *src_oa,
                 RETURN(rc);
         }
 
-        if (exp == NULL)
-                RETURN(-EINVAL);
-
         lov = &exp->exp_obd->u.lov;
-
         if (!lov->desc.ld_active_tgt_count)
                 RETURN(-EIO);
 
-        /* Recreate a specific object id at the given OST index */
-        if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
-            (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
-                 struct lov_stripe_md obj_md;
-                 struct lov_stripe_md *obj_mdp = &obj_md;
-
-                 ost_idx = src_oa->o_nlink;
-                 lsm = *ea;
-                 if (lsm == NULL)
-                        RETURN(-EINVAL);
-                 if (ost_idx >= lov->desc.ld_tgt_count)
-                         RETURN(-EINVAL);
-                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
-                         if (lsm->lsm_oinfo[i].loi_ost_idx == ost_idx) {
-                                 if (lsm->lsm_oinfo[i].loi_id != src_oa->o_id ||
-                                     lsm->lsm_oinfo[i].loi_gr != src_oa->o_gr) {
-                                         RETURN(-EINVAL);
-                                 }
-                                 break;
-                         }
-                 }
-                 if (i == lsm->lsm_stripe_count)
-                         RETURN(-EINVAL);
-
-                 rc = obd_create(lov->tgts[ost_idx].ltd_exp, src_oa,
-                                 &obj_mdp, oti);
-                 RETURN(rc);
-        }
-
-        ret_oa = obdo_alloc();
-        if (!ret_oa)
-                RETURN(-ENOMEM);
-
-        tmp_oa = obdo_alloc();
-        if (!tmp_oa)
-                GOTO(out_oa, rc = -ENOMEM);
-
-        lsm = *ea;
-        if (lsm == NULL) {
-                int stripes;
-                ost_count = lov_get_stripecnt(lov, 0);
-
-                /* If the MDS file was truncated up to some size, stripe over
-                 * enough OSTs to allow the file to be created at that size. */
-                if (src_oa->o_valid & OBD_MD_FLSIZE) {
-                        stripes=((src_oa->o_size+LUSTRE_STRIPE_MAXBYTES)>>12)-1;
-                        do_div(stripes, (__u32)(LUSTRE_STRIPE_MAXBYTES >> 12));
-
-                        if (stripes > lov->desc.ld_active_tgt_count)
-                                RETURN(-EFBIG);
-                        if (stripes < ost_count)
-                                stripes = ost_count;
-                } else {
-                        stripes = ost_count;
-                }
-
-                rc = lov_alloc_memmd(&lsm, stripes, lov->desc.ld_pattern ?
-                                     lov->desc.ld_pattern : LOV_PATTERN_RAID0);
-                if (rc < 0)
-                        GOTO(out_tmp, rc);
-
-                rc = 0;
-        }
-
-        ost_count = lov->desc.ld_tgt_count;
-
-        LASSERT(src_oa->o_gr > 0);
-        LASSERT(src_oa->o_valid & OBD_MD_FLID);
-        lsm->lsm_object_id = src_oa->o_id;
-        lsm->lsm_object_gr = src_oa->o_gr;
-        if (!lsm->lsm_stripe_size)
-                lsm->lsm_stripe_size = lov->desc.ld_default_stripe_size;
-        if (!lsm->lsm_pattern) {
-                lsm->lsm_pattern = lov->desc.ld_pattern ?
-                        lov->desc.ld_pattern : LOV_PATTERN_RAID0;
-        }
-
-        if (*ea == NULL || lsm->lsm_oinfo[0].loi_ost_idx >= ost_count) {
-                if (--ost_start_count <= 0) {
-                        ost_start_idx = ll_insecure_random_int();
-                        ost_start_count = LOV_CREATE_RESEED_INTERVAL;
-                } else if (lsm->lsm_stripe_count >=
-                           lov->desc.ld_active_tgt_count) {
-                        /* If we allocate from all of the stripes, make the
-                         * next file start on the next OST. */
-                        ++ost_start_idx;
-                }
-                ost_idx = ost_start_idx % ost_count;
-        } else {
-                ost_idx = lsm->lsm_oinfo[0].loi_ost_idx;
-        }
-
-        CDEBUG(D_INODE, "allocating %d subobjs for objid "LPX64" at idx %d\n",
-               lsm->lsm_stripe_count, lsm->lsm_object_id, ost_idx);
-
-        /* XXX LOV STACKING: need to figure out how many real OSCs */
-        if (oti && (src_oa->o_valid & OBD_MD_FLCOOKIE)) {
-                oti_alloc_cookies(oti, lsm->lsm_stripe_count);
-                if (!oti->oti_logcookies)
-                        GOTO(out_cleanup, rc = -ENOMEM);
-                cookies = oti->oti_logcookies;
-        }
-
-        loi = lsm->lsm_oinfo;
-        for (i = 0; i < ost_count; i++, ost_idx = (ost_idx + 1) % ost_count) {
-                struct lov_stripe_md obj_md;
-                struct lov_stripe_md *obj_mdp = &obj_md;
-                int err;
-
-                ++ost_start_idx;
-                if (lov->tgts[ost_idx].active == 0) {
-                        if (!obd_uuid_empty(&lov->tgts[ost_idx].uuid))
-                                CDEBUG(D_HA, "lov idx %d inactive\n", ost_idx);
-                        continue;
-                }
-
-                /* create data objects with "parent" OA */
-                memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
-
-                /* XXX When we start creating objects on demand, we need to
-                 *     make sure that we always create the object on the
-                 *     stripe which holds the existing file size.
-                 */
-                if (src_oa->o_valid & OBD_MD_FLSIZE) {
-                        if (lov_stripe_offset(lsm, src_oa->o_size, i,
-                                              &tmp_oa->o_size) < 0 &&
-                            tmp_oa->o_size)
-                                tmp_oa->o_size--;
-
-                        CDEBUG(D_INODE, "stripe %d has size "LPU64"/"LPU64"\n",
-                               i, tmp_oa->o_size, src_oa->o_size);
-                }
+        LASSERT(oti->oti_flags & OBD_MODE_CROW);
+                
+        /* main creation loop */
+        rc = lov_prep_create_set(exp, ea, src_oa, oti, &set);
+        if (rc)
+                RETURN(rc);
 
+        list_for_each (pos, &set->set_list) {
+                struct lov_request *req = 
+                        list_entry(pos, struct lov_request, rq_link);
 
                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
-                err = obd_create(lov->tgts[ost_idx].ltd_exp, tmp_oa, &obj_mdp,
-                                 oti);
-                if (err) {
-                        if (lov->tgts[ost_idx].active) {
-                                CERROR("error creating objid "LPX64" sub-object"
-                                       " on OST idx %d/%d: rc = %d\n",
-                                       src_oa->o_id, ost_idx,
-                                       lsm->lsm_stripe_count, err);
-                                if (err > 0) {
-                                        CERROR("obd_create returned invalid "
-                                               "err %d\n", err);
-                                        err = -EIO;
-                                }
-                        }
-                        if (!rc)
-                                rc = err;
-                        continue;
-                }
-                if (oti->oti_objid)
-                        oti->oti_objid[ost_idx] = tmp_oa->o_id;
-                loi->loi_id = tmp_oa->o_id;
-                loi->loi_gr = tmp_oa->o_gr;
-                loi->loi_ost_idx = ost_idx;
-                loi->loi_ost_gen = lov->tgts[ost_idx].ltd_gen;
-                CDEBUG(D_INODE, "objid "LPX64" has subobj "LPX64" at "
-                      "idx %d gen %d\n", lsm->lsm_object_id, loi->loi_id,
-                       ost_idx, loi->loi_ost_gen);
-
-                lov_merge_attrs(ret_oa, tmp_oa, tmp_oa->o_valid, lsm,
-                                obj_alloc, &set);
-                loi_init(loi);
-
-                if (cookies)
-                        ++oti->oti_logcookies;
-                if (tmp_oa->o_valid & OBD_MD_FLCOOKIE)
-                        ++cookie_sent;
-                ++obj_alloc;
-                ++loi;
-
-                /* If we have allocated enough objects, we are OK */
-                if (obj_alloc == lsm->lsm_stripe_count)
-                        GOTO(out_done, rc = 0);
-        }
-
-        if (obj_alloc == 0) {
-                if (rc == 0)
-                        rc = -EIO;
-                GOTO(out_cleanup, rc);
-        }
-
-        /* If we were passed specific striping params, then a failure to
-         * meet those requirements is an error, since we can't reallocate
-         * that memory (it might be part of a larger array or something).
-         *
-         * We can only get here if lsm_stripe_count was originally > 1.
-         */
-        if (*ea != NULL) {
-                CERROR("can't lstripe objid "LPX64": have %u want %u, rc %d\n",
-                       lsm->lsm_object_id, obj_alloc, lsm->lsm_stripe_count,rc);
-                if (rc == 0)
-                        rc = -EFBIG;
-                GOTO(out_cleanup, rc);
-        } else {
-                struct lov_stripe_md *lsm_new;
-                /* XXX LOV STACKING call into osc for sizes */
-                unsigned oldsize, newsize;
-
-                if (oti && cookies && cookie_sent) {
-                        oldsize = lsm->lsm_stripe_count * sizeof(*cookies);
-                        newsize = obj_alloc * sizeof(*cookies);
-
-                        oti_alloc_cookies(oti, obj_alloc);
-                        if (oti->oti_logcookies) {
-                                memcpy(oti->oti_logcookies, cookies, newsize);
-                                OBD_FREE(cookies, oldsize);
-                                cookies = oti->oti_logcookies;
-                        } else {
-                                CWARN("'leaking' %d bytes\n", oldsize-newsize);
-                        }
-                }
-
-                CWARN("using fewer stripes for object "LPX64": old %u new %u\n",
-                      lsm->lsm_object_id, lsm->lsm_stripe_count, obj_alloc);
-                oldsize = lov_stripe_md_size(lsm->lsm_stripe_count);
-                newsize = lov_stripe_md_size(obj_alloc);
-                OBD_ALLOC(lsm_new, newsize);
-                if (lsm_new != NULL) {
-                        memcpy(lsm_new, lsm, newsize);
-                        lsm_new->lsm_stripe_count = obj_alloc;
-                        OBD_FREE(lsm, oldsize);
-                        lsm = lsm_new;
-                } else {
-                        CWARN("'leaking' %d bytes\n", oldsize - newsize);
-                }
-                rc = 0;
-        }
-        EXIT;
- out_done:
-        *ea = lsm;
-        if (src_oa->o_valid & OBD_MD_FLSIZE &&
-            ret_oa->o_size != src_oa->o_size) {
-                CERROR("original size "LPU64" isn't new object size "LPU64"\n",
-                       src_oa->o_size, ret_oa->o_size);
-                LBUG();
-        }
-        ret_oa->o_id = src_oa->o_id;
-        ret_oa->o_gr = src_oa->o_gr;
-        ret_oa->o_valid |= OBD_MD_FLGROUP;
-        memcpy(src_oa, ret_oa, sizeof(*src_oa));
-
- out_tmp:
-        obdo_free(tmp_oa);
- out_oa:
-        obdo_free(ret_oa);
-        if (oti && cookies) {
-                oti->oti_logcookies = cookies;
-                if (!cookie_sent) {
-                        oti_free_cookies(oti);
-                        src_oa->o_valid &= ~OBD_MD_FLCOOKIE;
-                } else {
-                        src_oa->o_valid |= OBD_MD_FLCOOKIE;
-                }
+                rc = obd_create(lov->tgts[req->rq_idx].ltd_exp,
+                                req->rq_oa, NULL, 0, &req->rq_md, oti);
+                lov_update_create_set(set, req, rc);
         }
+        rc = lov_fini_create_set(set, ea);
         RETURN(rc);
-
- out_cleanup:
-        while (obj_alloc-- > 0) {
-                struct obd_export *sub_exp;
-                int err;
-
-                --loi;
-                sub_exp = lov->tgts[loi->loi_ost_idx].ltd_exp;
-                /* destroy already created objects here */
-                memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
-                tmp_oa->o_id = loi->loi_id;
-
-                err = obd_destroy(sub_exp, tmp_oa, NULL, oti);
-                if (err)
-                        CERROR("Failed to uncreate objid "LPX64" subobj "LPX64
-                               " on OST idx %d: rc = %d\n", src_oa->o_id,
-                               loi->loi_id, loi->loi_ost_idx, err);
-        }
-        if (*ea == NULL)
-                obd_free_memmd(exp, &lsm);
-        goto out_tmp;
-}
-
-static int lov_revalidate_policy(struct lov_obd *lov, struct lov_stripe_md *lsm)
-{
-        static int next_idx = 0;
-        struct lov_tgt_desc *tgt;
-        int i, count;
-
-        /* XXX - we should do something clever and take lsm
-         * into account but just do round robin for now. */
-
-        /* last_idx must always be less that count because
-         * ld_tgt_count currently cannot shrink. */
-        count = lov->desc.ld_tgt_count;
-
-        for (i = next_idx, tgt = lov->tgts + i; i < count; i++, tgt++) {
-                if (tgt->active) {
-                        next_idx = (i + 1) % count;
-                        RETURN(i);
-                }
-        }
-
-        for (i = 0, tgt = lov->tgts; i < next_idx; i++, tgt++) {
-                if (tgt->active) {
-                        next_idx = (i + 1) % count;
-                        RETURN(i);
-                }
-        }
-
-        RETURN(-EIO);
 }
 
 #define lsm_bad_magic(LSMP)                                     \
@@ -1166,10 +790,11 @@ static int lov_revalidate_policy(struct lov_obd *lov, struct lov_stripe_md *lsm)
 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
 {
-        struct obdo tmp;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i;
+        int rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
@@ -1179,40 +804,40 @@ static int lov_destroy(struct obd_export *exp, struct obdo *oa,
                 RETURN(-ENODEV);
 
         lov = &exp->exp_obd->u.lov;
-        loi = lsm->lsm_oinfo;
-        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
-                int err;
+        rc = lov_prep_destroy_set(exp, oa, lsm, oti, &set);
+        if (rc)
+                RETURN(rc);
 
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        /* Orphan clean up will (someday) fix this up. */
-                        if (oti != NULL && oa->o_valid & OBD_MD_FLCOOKIE)
-                                oti->oti_logcookies++;
-                        continue;
-                }
+        list_for_each (pos, &set->set_list) {
+                int err;
+                req = list_entry(pos, struct lov_request, rq_link);
 
-                memcpy(&tmp, oa, sizeof(tmp));
-                tmp.o_id = loi->loi_id;
-                err = obd_destroy(lov->tgts[loi->loi_ost_idx].ltd_exp, &tmp,
-                                  NULL, oti);
-                if (err && lov->tgts[loi->loi_ost_idx].active) {
-                        CDEBUG(D_INODE, "error: destroying objid "LPX64" subobj "
-                               LPX64" on OST idx %d: rc = %d\n",
-                               oa->o_id, loi->loi_id, loi->loi_ost_idx, err);
+                /* XXX update the cookie position */
+                oti->oti_logcookies = set->set_cookies + req->rq_stripe;
+                rc = obd_destroy(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
+                                 NULL, oti);
+                err = lov_update_common_set(set, req, rc);
+                if (rc) {
+                        CERROR("error: destroying objid "LPX64" subobj "
+                               LPX64" on OST idx %d: rc = %d\n", 
+                               set->set_oa->o_id, req->rq_oa->o_id, 
+                               req->rq_idx, rc);
                         if (!rc)
                                 rc = err;
                 }
         }
+        lov_fini_destroy_set(set);
         RETURN(rc);
 }
 
 static int lov_getattr(struct obd_export *exp, struct obdo *oa,
                        struct lov_stripe_md *lsm)
 {
-        struct obdo tmp;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int i, rc = 0, set = 0;
+        int err = 0, rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
@@ -1222,73 +847,49 @@ static int lov_getattr(struct obd_export *exp, struct obdo *oa,
                 RETURN(-ENODEV);
 
         lov = &exp->exp_obd->u.lov;
+        
+        rc = lov_prep_getattr_set(exp, oa, lsm, &set);
+        if (rc)
+                RETURN(rc);
 
-        CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
-               lsm->lsm_object_id, lsm->lsm_stripe_count, lsm->lsm_stripe_size);
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
-
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                
                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
-                       "%u\n", oa->o_id, i, loi->loi_id, loi->loi_ost_idx);
-                /* create data objects with "parent" OA */
-                memcpy(&tmp, oa, sizeof(tmp));
-                tmp.o_id = loi->loi_id;
+                       "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id, 
+                       req->rq_idx);
 
-                err = obd_getattr(lov->tgts[loi->loi_ost_idx].ltd_exp, &tmp,
-                                  NULL);
+                rc = obd_getattr(lov->tgts[req->rq_idx].ltd_exp, 
+                                 req->rq_oa, NULL);
+                err = lov_update_common_set(set, req, rc);
                 if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: getattr objid "LPX64" subobj "
-                                       LPX64" on OST idx %d: rc = %d\n",
-                                       oa->o_id, loi->loi_id, loi->loi_ost_idx,
-                                       err);
-                                RETURN(err);
-                        }
-                } else {
-                        lov_merge_attrs(oa, &tmp, tmp.o_valid, lsm, i, &set);
+                        CERROR("error: getattr objid "LPX64" subobj "
+                               LPX64" on OST idx %d: rc = %d\n",
+                               set->set_oa->o_id, req->rq_oa->o_id, 
+                               req->rq_idx, err);
+                        break;
                 }
         }
-        if (!set)
-                rc = -EIO;
+        
+        rc = lov_fini_getattr_set(set);
+        if (err)
+                rc = err;
         RETURN(rc);
 }
 
 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset, void *data,
                                  int rc)
 {
-        struct lov_getattr_async_args *aa = data;
-        struct lov_stripe_md *lsm = aa->aa_lsm;
-        struct obdo          *oa = aa->aa_oa;
-        struct obdo          *obdos = aa->aa_obdos;
-        struct lov_oinfo     *loi;
-        int                   i;
-        int                   set = 0;
+        struct lov_request_set *lovset = (struct lov_request_set *)data;
         ENTRY;
 
-        if (rc == 0) {
-                /* NB all stripe requests succeeded to get here */
-
-                loi = lsm->lsm_oinfo;
-                for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
-                        if (obdos[i].o_valid == 0)      /* inactive stripe */
-                                continue;
-
-                        lov_merge_attrs(oa, &obdos[i], obdos[i].o_valid, lsm,
-                                        i, &set);
-                }
-
-                if (!set) {
-                        CERROR ("No stripes had valid attrs\n");
-                        rc = -EIO;
-                }
+        /* don't do attribute merge if this aysnc op failed */
+        if (rc) {
+                lovset->set_completes = 0;
+                lov_fini_getattr_set(lovset);
+        } else {
+                rc = lov_fini_getattr_set(lovset);
         }
-
-        OBD_FREE (obdos, lsm->lsm_stripe_count * sizeof (*obdos));
         RETURN (rc);
 }
 
@@ -1296,11 +897,11 @@ static int lov_getattr_async(struct obd_export *exp, struct obdo *oa,
                               struct lov_stripe_md *lsm,
                               struct ptlrpc_request_set *rqset)
 {
-        struct obdo *obdos;
+        struct lov_request_set *lovset;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        struct lov_getattr_async_args *aa;
-        int i, rc = 0, set = 0;
+        struct list_head *pos;
+        struct lov_request *req;
+        int rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
@@ -1311,67 +912,50 @@ static int lov_getattr_async(struct obd_export *exp, struct obdo *oa,
 
         lov = &exp->exp_obd->u.lov;
 
-        OBD_ALLOC (obdos, lsm->lsm_stripe_count * sizeof (*obdos));
-        if (obdos == NULL)
-                RETURN(-ENOMEM);
+        rc = lov_prep_getattr_set(exp, oa, lsm, &lovset);
+        if (rc)
+                RETURN(rc);
 
         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
                lsm->lsm_object_id, lsm->lsm_stripe_count, lsm->lsm_stripe_size);
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        /* leaves obdos[i].obd_valid unset */
-                        continue;
-                }
-
-                CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at "
-                       "idx %u gen %d\n", oa->o_id, i, loi->loi_id,
-                       loi->loi_ost_idx, loi->loi_ost_gen);
 
-                /* create data objects with "parent" OA */
-                memcpy(&obdos[i], oa, sizeof(obdos[i]));
-                obdos[i].o_id = loi->loi_id;
-
-                err = obd_getattr_async(lov->tgts[loi->loi_ost_idx].ltd_exp,
-                                         &obdos[i], NULL, rqset);
-                if (err) {
+        list_for_each (pos, &lovset->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                
+                CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
+                       "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id, 
+                       req->rq_idx);
+                rc = obd_getattr_async(lov->tgts[req->rq_idx].ltd_exp,
+                                       req->rq_oa, NULL, rqset);
+                if (rc) {
                         CERROR("error: getattr objid "LPX64" subobj "
                                LPX64" on OST idx %d: rc = %d\n",
-                               oa->o_id, loi->loi_id, loi->loi_ost_idx,
-                               err);
-                        GOTO(out_obdos, rc = err);
+                               lovset->set_oa->o_id, req->rq_oa->o_id, 
+                               req->rq_idx, rc);
+                        GOTO(out, rc);
                 }
-                set = 1;
+                lov_update_common_set(lovset, req, rc);
         }
-        if (!set)
-                GOTO (out_obdos, rc = -EIO);
-
+        
+        LASSERT(rc == 0);
         LASSERT (rqset->set_interpret == NULL);
         rqset->set_interpret = lov_getattr_interpret;
-        LASSERT (sizeof (rqset->set_args) >= sizeof (*aa));
-        aa = (struct lov_getattr_async_args *)&rqset->set_args;
-        aa->aa_lsm = lsm;
-        aa->aa_oa = oa;
-        aa->aa_obdos = obdos;
-        aa->aa_lov = lov;
-        GOTO(out, rc = 0);
-
-out_obdos:
-        OBD_FREE (obdos, lsm->lsm_stripe_count * sizeof (*obdos));
+        rqset->set_arg = (void *)lovset;
+        RETURN(rc);
 out:
+        LASSERT(rc);
+        lov_fini_getattr_set(lovset);
         RETURN(rc);
 }
 
-
 static int lov_setattr(struct obd_export *exp, struct obdo *src_oa,
                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
 {
-        struct obdo *tmp_oa, *ret_oa;
+        struct lov_request_set *set;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i, set = 0;
+        struct list_head *pos;
+        struct lov_request *req;
+        int err = 0, rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
@@ -1380,266 +964,68 @@ static int lov_setattr(struct obd_export *exp, struct obdo *src_oa,
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
-        /* for now, we only expect time updates here */
         LASSERT(!(src_oa->o_valid & ~(OBD_MD_FLID|OBD_MD_FLTYPE | OBD_MD_FLMODE|
                                       OBD_MD_FLATIME | OBD_MD_FLMTIME |
                                       OBD_MD_FLCTIME | OBD_MD_FLFLAGS |
-                                      OBD_MD_FLSIZE | OBD_MD_FLGROUP)));
+                                      OBD_MD_FLSIZE | OBD_MD_FLGROUP |
+                                      OBD_MD_FLUID | OBD_MD_FLGID)));
 
         LASSERT(!(src_oa->o_valid & OBD_MD_FLGROUP) || src_oa->o_gr > 0);
 
-        ret_oa = obdo_alloc();
-        if (!ret_oa)
-                RETURN(-ENOMEM);
-
-        tmp_oa = obdo_alloc();
-        if (!tmp_oa)
-                GOTO(out_oa, rc = -ENOMEM);
-
         lov = &exp->exp_obd->u.lov;
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
-
-                memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
-                tmp_oa->o_id = loi->loi_id;
-                LASSERT(!(tmp_oa->o_valid & OBD_MD_FLGROUP) || tmp_oa->o_gr>0);
-
-                if (src_oa->o_valid & OBD_MD_FLSIZE) {
-                        if (lov_stripe_offset(lsm, src_oa->o_size, i,
-                                              &tmp_oa->o_size) < 0 &&
-                            tmp_oa->o_size)
-                                tmp_oa->o_size--;
-
-                        CDEBUG(D_INODE, "stripe %d has size "LPU64"/"LPU64"\n",
-                               i, tmp_oa->o_size, src_oa->o_size);
-                }
+        rc = lov_prep_setattr_set(exp, src_oa, lsm, NULL, &set);
+        if (rc)
+                RETURN(rc);
 
-                err = obd_setattr(lov->tgts[loi->loi_ost_idx].ltd_exp, tmp_oa,
-                                  NULL, NULL);
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                
+                rc = obd_setattr(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
+                                 NULL, NULL);
+                err = lov_update_common_set(set, req, rc);
                 if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: setattr objid "LPX64" subobj "
-                                       LPX64" on OST idx %d: rc = %d\n",
-                                       src_oa->o_id, loi->loi_id,
-                                       loi->loi_ost_idx, err);
-                                if (!rc)
-                                        rc = err;
-                        }
-                        continue;
+                        CERROR("error: setattr objid "LPX64" subobj "
+                               LPX64" on OST idx %d: rc = %d\n",
+                               set->set_oa->o_id, req->rq_oa->o_id,
+                               req->rq_idx, err);
+                        if (!rc)
+                                rc = err;
                 }
-                lov_merge_attrs(ret_oa, tmp_oa, tmp_oa->o_valid, lsm, i, &set);
         }
-        if (!set && !rc)
-                rc = -EIO;
-
-        ret_oa->o_id = src_oa->o_id;
-        memcpy(src_oa, ret_oa, sizeof(*src_oa));
-        GOTO(out_tmp, rc);
-out_tmp:
-        obdo_free(tmp_oa);
-out_oa:
-        obdo_free(ret_oa);
-        return rc;
+        err = lov_fini_setattr_set(set);
+        if (!rc)
+                rc = err;
+        RETURN(rc);
 }
 
-/* we have an offset in file backed by an lov and want to find out where
- * that offset lands in our given stripe of the file.  for the easy
- * case where the offset is within the stripe, we just have to scale the
- * offset down to make it relative to the stripe instead of the lov.
- *
- * the harder case is what to do when the offset doesn't intersect the
- * stripe.  callers will want start offsets clamped ahead to the start
- * of the nearest stripe in the file.  end offsets similarly clamped to the
- * nearest ending byte of a stripe in the file:
- *
- * all this function does is move offsets to the nearest region of the
- * stripe, and it does its work "mod" the full length of all the stripes.
- * consider a file with 3 stripes:
- *
- *             S                                              E
- * ---------------------------------------------------------------------
- * |    0    |     1     |     2     |    0    |     1     |     2     |
- * ---------------------------------------------------------------------
- *
- * to find stripe 1's offsets for S and E, it divides by the full stripe
- * width and does its math in the context of a single set of stripes:
- *
- *             S         E
- * -----------------------------------
- * |    0    |     1     |     2     |
- * -----------------------------------
- *
- * it'll notice that E is outside stripe 1 and clamp it to the end of the
- * stripe, then multiply it back out by lov_off to give the real offsets in
- * the stripe:
- *
- *   S                   E
- * ---------------------------------------------------------------------
- * |    1    |     1     |     1     |    1    |     1     |     1     |
- * ---------------------------------------------------------------------
- *
- * it would have done similarly and pulled S forward to the start of a 1
- * stripe if, say, S had landed in a 0 stripe.
- *
- * this rounding isn't always correct.  consider an E lov offset that lands
- * on a 0 stripe, the "mod stripe width" math will pull it forward to the
- * start of a 1 stripe, when in fact it wanted to be rounded back to the end
- * of a previous 1 stripe.  this logic is handled by callers and this is why:
- *
- * this function returns < 0 when the offset was "before" the stripe and
- * was moved forward to the start of the stripe in question;  0 when it
- * falls in the stripe and no shifting was done; > 0 when the offset
- * was outside the stripe and was pulled back to its final byte. */
-static int lov_stripe_offset(struct lov_stripe_md *lsm, obd_off lov_off,
-                             int stripeno, obd_off *obd_off)
+static int lov_revalidate_policy(struct lov_obd *lov, struct lov_stripe_md *lsm)
 {
-        unsigned long ssize  = lsm->lsm_stripe_size;
-        unsigned long swidth = ssize * lsm->lsm_stripe_count;
-        unsigned long stripe_off, this_stripe;
-        int ret = 0;
-
-        if (lov_off == OBD_OBJECT_EOF) {
-                *obd_off = OBD_OBJECT_EOF;
-                return 0;
-        }
+        static int next_idx = 0;
+        struct lov_tgt_desc *tgt;
+        int i, count;
 
-        /* do_div(a, b) returns a % b, and a = a / b */
-        stripe_off = do_div(lov_off, swidth);
+        /* XXX - we should do something clever and take lsm
+         * into account but just do round robin for now. */
 
-        this_stripe = stripeno * ssize;
-        if (stripe_off < this_stripe) {
-                stripe_off = 0;
-                ret = -1;
-        } else {
-                stripe_off -= this_stripe;
+        /* last_idx must always be less that count because
+         * ld_tgt_count currently cannot shrink. */
+        count = lov->desc.ld_tgt_count;
 
-                if (stripe_off >= ssize) {
-                        stripe_off = ssize;
-                        ret = 1;
+        for (i = next_idx, tgt = lov->tgts + i; i < count; i++, tgt++) {
+                if (tgt->active) {
+                        next_idx = (i + 1) % count;
+                        RETURN(i);
                 }
         }
 
-        *obd_off = lov_off * ssize + stripe_off;
-        return ret;
-}
-
-/* Given a whole-file size and a stripe number, give the file size which
- * corresponds to the individual object of that stripe.
- *
- * This behaves basically in the same was as lov_stripe_offset, except that
- * file sizes falling before the beginning of a stripe are clamped to the end
- * of the previous stripe, not the beginning of the next:
- *
- *                                               S
- * ---------------------------------------------------------------------
- * |    0    |     1     |     2     |    0    |     1     |     2     |
- * ---------------------------------------------------------------------
- *
- * if clamped to stripe 2 becomes:
- *
- *                                   S
- * ---------------------------------------------------------------------
- * |    0    |     1     |     2     |    0    |     1     |     2     |
- * ---------------------------------------------------------------------
- */
-static obd_off lov_size_to_stripe(struct lov_stripe_md *lsm, obd_off file_size,
-                                  int stripeno)
-{
-        unsigned long ssize  = lsm->lsm_stripe_size;
-        unsigned long swidth = ssize * lsm->lsm_stripe_count;
-        unsigned long stripe_off, this_stripe;
-
-        if (file_size == OBD_OBJECT_EOF)
-                return OBD_OBJECT_EOF;
-
-        /* do_div(a, b) returns a % b, and a = a / b */
-        stripe_off = do_div(file_size, swidth);
-
-        this_stripe = stripeno * ssize;
-        if (stripe_off < this_stripe) {
-                /* Move to end of previous stripe, or zero */
-                if (file_size > 0) {
-                        file_size--;
-                        stripe_off = ssize;
-                } else {
-                        stripe_off = 0;
-                }
-        } else {
-                stripe_off -= this_stripe;
-
-                if (stripe_off >= ssize) {
-                        /* Clamp to end of this stripe */
-                        stripe_off = ssize;
+        for (i = 0, tgt = lov->tgts; i < next_idx; i++, tgt++) {
+                if (tgt->active) {
+                        next_idx = (i + 1) % count;
+                        RETURN(i);
                 }
         }
 
-        return (file_size * ssize + stripe_off);
-}
-
-/* given an extent in an lov and a stripe, calculate the extent of the stripe
- * that is contained within the lov extent.  this returns true if the given
- * stripe does intersect with the lov extent. */
-static int lov_stripe_intersects(struct lov_stripe_md *lsm, int stripeno,
-                                 obd_off start, obd_off end,
-                                 obd_off *obd_start, obd_off *obd_end)
-{
-        int start_side = 0, end_side = 0;
-
-        switch (lsm->lsm_pattern) {
-        case LOV_PATTERN_RAID0:
-                start_side = lov_stripe_offset(lsm, start, stripeno, obd_start);
-                end_side = lov_stripe_offset(lsm, end, stripeno, obd_end);
-                break;
-        case LOV_PATTERN_CMOBD:
-                *obd_start = start;
-                *obd_end = end;
-                start_side = end_side = 0;
-                break;
-        default:
-                LBUG();
-        }
-
-        CDEBUG(D_INODE, "["LPU64"->"LPU64"] -> [(%d) "LPU64"->"LPU64" (%d)]\n",
-               start, end, start_side, *obd_start, *obd_end, end_side);
-
-        /* this stripe doesn't intersect the file extent when neither
-         * start or the end intersected the stripe and obd_start and
-         * obd_end got rounded up to the save value. */
-        if (start_side != 0 && end_side != 0 && *obd_start == *obd_end)
-                return 0;
-
-        /* as mentioned in the lov_stripe_offset commentary, end
-         * might have been shifted in the wrong direction.  This
-         * happens when an end offset is before the stripe when viewed
-         * through the "mod stripe size" math. we detect it being shifted
-         * in the wrong direction and touch it up.
-         * interestingly, this can't underflow since end must be > start
-         * if we passed through the previous check.
-         * (should we assert for that somewhere?) */
-        if (end_side != 0)
-                (*obd_end)--;
-
-        return 1;
-}
-
-/* compute which stripe number "lov_off" will be written into */
-static int lov_stripe_number(struct lov_stripe_md *lsm, obd_off lov_off)
-{
-        unsigned long ssize  = lsm->lsm_stripe_size;
-        unsigned long swidth = ssize * lsm->lsm_stripe_count;
-        unsigned long stripe_off;
-
-        if (lsm->lsm_pattern == LOV_PATTERN_CMOBD)
-                return 0;
-
-        stripe_off = do_div(lov_off, swidth);
-
-        return stripe_off / ssize;
+        RETURN(-EIO);
 }
 
 static int lov_revalidate_md(struct obd_export *exp, struct obdo *src_oa,
@@ -1678,7 +1064,7 @@ static int lov_revalidate_md(struct obd_export *exp, struct obdo *src_oa,
                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
                 osc_exp = lov->tgts[ost_idx].ltd_exp;
-                rc = obd_create(osc_exp, tmp_oa, &obj_mdp, oti);
+                rc = obd_create(osc_exp, tmp_oa, NULL, 0, &obj_mdp, oti);
                 if (rc) {
                         CERROR("error creating new subobj at idx %d; "
                                "rc = %d\n", ost_idx, rc);
@@ -1687,6 +1073,7 @@ static int lov_revalidate_md(struct obd_export *exp, struct obdo *src_oa,
                 if (oti->oti_objid)
                         oti->oti_objid[ost_idx] = tmp_oa->o_id;
                 loi->loi_id = tmp_oa->o_id;
+                loi->loi_gr = tmp_oa->o_gr;
                 loi->loi_ost_idx = ost_idx;
                 loi->loi_ost_gen = lov->tgts[ost_idx].ltd_gen;
                 CDEBUG(D_INODE, "replacing objid "LPX64" subobj "LPX64
@@ -1710,10 +1097,11 @@ static int lov_punch(struct obd_export *exp, struct obdo *oa,
                      struct lov_stripe_md *lsm,
                      obd_off start, obd_off end, struct obd_trans_info *oti)
 {
-        struct obdo tmp;
+        struct lov_request_set *set;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i;
+        struct list_head *pos;
+        struct lov_request *req;
+        int err = 0, rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
@@ -1723,46 +1111,39 @@ static int lov_punch(struct obd_export *exp, struct obdo *oa,
                 RETURN(-ENODEV);
 
         lov = &exp->exp_obd->u.lov;
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                obd_off starti, endi;
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
-
-                if (!lov_stripe_intersects(lsm, i, start, end, &starti, &endi))
-                        continue;
+        rc = lov_prep_punch_set(exp, oa, lsm, start, end, oti, &set);
+        if (rc)
+                RETURN(rc);
 
-                /* create data objects with "parent" OA */
-                memcpy(&tmp, oa, sizeof(tmp));
-                tmp.o_id = loi->loi_id;
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
 
-                err = obd_punch(lov->tgts[loi->loi_ost_idx].ltd_exp, &tmp, NULL,
-                                starti, endi, NULL);
+                rc = obd_punch(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa, 
+                               NULL, req->rq_extent.start, 
+                               req->rq_extent.end, NULL);
+                err = lov_update_punch_set(set, req, rc);
                 if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: punch objid "LPX64" subobj "LPX64
-                                       " on OST idx %d: rc = %d\n", oa->o_id,
-                                       loi->loi_id, loi->loi_ost_idx, err);
-                        }
+                        CERROR("error: punch objid "LPX64" subobj "LPX64
+                               " on OST idx %d: rc = %d\n", set->set_oa->o_id,
+                               req->rq_oa->o_id, req->rq_idx, rc);
                         if (!rc)
                                 rc = err;
-                } else {
-                        loi->loi_kms = loi->loi_rss = starti;
                 }
         }
+        err = lov_fini_punch_set(set);
+        if (!rc)
+                rc = err;
         RETURN(rc);
 }
 
 static int lov_sync(struct obd_export *exp, struct obdo *oa,
                     struct lov_stripe_md *lsm, obd_off start, obd_off end)
 {
-        struct obdo *tmp;
+        struct lov_request_set *set;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i;
+        struct list_head *pos;
+        struct lov_request *req;
+        int err = 0, rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
@@ -1771,40 +1152,28 @@ static int lov_sync(struct obd_export *exp, struct obdo *oa,
         if (!exp->exp_obd)
                 RETURN(-ENODEV);
 
-        tmp = obdo_alloc();
-        if (!tmp)
-                RETURN(-ENOMEM);
-
         lov = &exp->exp_obd->u.lov;
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                obd_off starti, endi;
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
-
-                if (!lov_stripe_intersects(lsm, i, start, end, &starti, &endi))
-                        continue;
+        rc = lov_prep_sync_set(exp, oa, lsm, start, end, &set);
+        if (rc)
+                RETURN(rc);
 
-                memcpy(tmp, oa, sizeof(*tmp));
-                tmp->o_id = loi->loi_id;
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
 
-                err = obd_sync(lov->tgts[loi->loi_ost_idx].ltd_exp, tmp, NULL,
-                               starti, endi);
+                rc = obd_sync(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa, 
+                              NULL, req->rq_extent.start, req->rq_extent.end);
+                err = lov_update_common_set(set, req, rc);
                 if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: fsync objid "LPX64" subobj "LPX64
-                                       " on OST idx %d: rc = %d\n", oa->o_id,
-                                       loi->loi_id, loi->loi_ost_idx, err);
-                        }
+                        CERROR("error: fsync objid "LPX64" subobj "LPX64
+                               " on OST idx %d: rc = %d\n", set->set_oa->o_id,
+                               req->rq_oa->o_id, req->rq_idx, rc);
                         if (!rc)
                                 rc = err;
                 }
         }
-
-        obdo_free(tmp);
+        err = lov_fini_sync_set(set);
+        if (!rc)
+                rc = err;
         RETURN(rc);
 }
 
@@ -1843,155 +1212,58 @@ static int lov_brw(int cmd, struct obd_export *exp, struct obdo *src_oa,
                    struct lov_stripe_md *lsm, obd_count oa_bufs,
                    struct brw_page *pga, struct obd_trans_info *oti)
 {
-        struct {
-                int bufct;
-                int index;
-                int subcount;
-                struct lov_stripe_md lsm;
-                int ost_idx;
-        } *stripeinfo, *si, *si_last;
-        struct obdo *ret_oa = NULL, *tmp_oa = NULL;
-        struct lov_obd *lov;
-        struct brw_page *ioarr;
-        struct lov_oinfo *loi;
-        int rc = 0, i, *where, stripe_count = lsm->lsm_stripe_count, set = 0;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int err, rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
                 RETURN(-EINVAL);
 
-        lov = &exp->exp_obd->u.lov;
-
         if (cmd == OBD_BRW_CHECK) {
                 rc = lov_brw_check(lov, src_oa, lsm, oa_bufs, pga);
                 RETURN(rc);
         }
 
-        OBD_ALLOC(stripeinfo, stripe_count * sizeof(*stripeinfo));
-        if (!stripeinfo)
-                RETURN(-ENOMEM);
-
-        OBD_ALLOC(where, sizeof(*where) * oa_bufs);
-        if (!where)
-                GOTO(out_sinfo, rc = -ENOMEM);
-
-        OBD_ALLOC(ioarr, sizeof(*ioarr) * oa_bufs);
-        if (!ioarr)
-                GOTO(out_where, rc = -ENOMEM);
-
-        if (src_oa) {
-                ret_oa = obdo_alloc();
-                if (!ret_oa)
-                        GOTO(out_ioarr, rc = -ENOMEM);
-
-                tmp_oa = obdo_alloc();
-                if (!tmp_oa)
-                        GOTO(out_oa, rc = -ENOMEM);
-        }
-
-        for (i = 0; i < oa_bufs; i++) {
-                where[i] = lov_stripe_number(lsm, pga[i].disk_offset);
-                stripeinfo[where[i]].bufct++;
-        }
-
-        for (i = 0, loi = lsm->lsm_oinfo, si_last = si = stripeinfo;
-             i < stripe_count; i++, loi++, si_last = si, si++) {
-                if (i > 0)
-                        si->index = si_last->index + si_last->bufct;
-                si->lsm.lsm_object_id = loi->loi_id;
-                si->lsm.lsm_object_gr = lsm->lsm_object_gr;
-                si->ost_idx = loi->loi_ost_idx;
-        }
-
-        for (i = 0; i < oa_bufs; i++) {
-                int which = where[i];
-                int shift;
-
-                shift = stripeinfo[which].index + stripeinfo[which].subcount;
-                LASSERT(shift < oa_bufs);
-                ioarr[shift] = pga[i];
-                lov_stripe_offset(lsm, pga[i].disk_offset, which,
-                                  &ioarr[shift].disk_offset);
-                stripeinfo[which].subcount++;
-        }
-
-        for (i = 0, si = stripeinfo; i < stripe_count; i++, si++) {
-                int shift = si->index;
-
-                if (lov->tgts[si->ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", si->ost_idx);
-                        GOTO(out_oa, rc = -EIO);
-                }
-
-                if (si->bufct) {
-                        LASSERT(shift < oa_bufs);
-                        if (src_oa)
-                                memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
-
-                        tmp_oa->o_id = si->lsm.lsm_object_id;
-                        rc = obd_brw(cmd, lov->tgts[si->ost_idx].ltd_exp,
-                                     tmp_oa, &si->lsm, si->bufct,
-                                     &ioarr[shift], oti);
-                        if (rc)
-                                GOTO(out_oa, rc);
+        rc = lov_prep_brw_set(exp, src_oa, lsm, oa_bufs, pga, oti, &set);
+        if (rc)
+                RETURN(rc);
 
-                        lov_merge_attrs(ret_oa, tmp_oa, tmp_oa->o_valid, lsm,
-                                        i, &set);
-                }
+        list_for_each (pos, &set->set_list) {
+                struct obd_export *sub_exp;
+                struct brw_page *sub_pga;
+                req = list_entry(pos, struct lov_request, rq_link);
+                
+                sub_exp = lov->tgts[req->rq_idx].ltd_exp;
+                sub_pga = set->set_pga + req->rq_pgaidx;
+                rc = obd_brw(cmd, sub_exp, req->rq_oa, req->rq_md, 
+                             req->rq_oabufs, sub_pga, oti);
+                if (rc)
+                        break;
+                lov_update_common_set(set, req, rc);
         }
 
-        ret_oa->o_id = src_oa->o_id;
-        memcpy(src_oa, ret_oa, sizeof(*src_oa));
-
-        GOTO(out_oa, rc);
- out_oa:
-        if (tmp_oa)
-                obdo_free(tmp_oa);
-        if (ret_oa)
-                obdo_free(ret_oa);
- out_ioarr:
-        OBD_FREE(ioarr, sizeof(*ioarr) * oa_bufs);
- out_where:
-        OBD_FREE(where, sizeof(*where) * oa_bufs);
- out_sinfo:
-        OBD_FREE(stripeinfo, stripe_count * sizeof(*stripeinfo));
-        return rc;
+        err = lov_fini_brw_set(set);
+        if (!rc)
+                rc = err;
+        RETURN(rc);
 }
 
-static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
-                             int rc)
-{
-        struct lov_brw_async_args *aa = data;
-        struct lov_stripe_md *lsm = aa->aa_lsm;
-        obd_count             oa_bufs = aa->aa_oa_bufs;
-        struct obdo          *oa = aa->aa_oa;
-        struct obdo          *obdos = aa->aa_obdos;
-        struct brw_page      *ioarr = aa->aa_ioarr;
-        struct lov_oinfo     *loi;
-        int i, set = 0;
-        ENTRY;
-
-        if (rc == 0) {
-                /* NB all stripe requests succeeded to get here */
-
-                for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-                     i++, loi++) {
-                        if (obdos[i].o_valid == 0)      /* inactive stripe */
-                                continue;
-
-                        lov_merge_attrs(oa, &obdos[i], obdos[i].o_valid, lsm,
-                                        i, &set);
-                }
-
-                if (!set) {
-                        CERROR("No stripes had valid attrs\n");
-                        rc = -EIO;
-                }
-        }
-        oa->o_id = lsm->lsm_object_id;
-
-        OBD_FREE(obdos, lsm->lsm_stripe_count * sizeof(*obdos));
-        OBD_FREE(ioarr, sizeof(*ioarr) * oa_bufs);
+static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
+                             int rc)
+{
+        struct lov_request_set *lovset = (struct lov_request_set *)data;
+        ENTRY;
+        
+        if (rc) {
+                lovset->set_completes = 0;
+                lov_fini_brw_set(lovset);
+        } else {
+                rc = lov_fini_brw_set(lovset);
+        }
+                
         RETURN(rc);
 }
 
@@ -2000,155 +1272,65 @@ static int lov_brw_async(int cmd, struct obd_export *exp, struct obdo *oa,
                          struct brw_page *pga, struct ptlrpc_request_set *set,
                          struct obd_trans_info *oti)
 {
-        struct {
-                int bufct;
-                int index;
-                int subcount;
-                struct lov_stripe_md lsm;
-                int ost_idx;
-        } *stripeinfo, *si, *si_last;
-        struct lov_obd *lov;
-        struct brw_page *ioarr;
-        struct obdo *obdos = NULL;
-        struct lov_oinfo *loi;
-        struct lov_brw_async_args *aa;
-        int rc = 0, i, *where, stripe_count = lsm->lsm_stripe_count;
+        struct lov_request_set *lovset;
+        struct lov_request *req;
+        struct list_head *pos;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
                 RETURN(-EINVAL);
 
-        lov = &exp->exp_obd->u.lov;
-
         if (cmd == OBD_BRW_CHECK) {
                 rc = lov_brw_check(lov, oa, lsm, oa_bufs, pga);
                 RETURN(rc);
         }
 
-        OBD_ALLOC(stripeinfo, stripe_count * sizeof(*stripeinfo));
-        if (!stripeinfo)
-                RETURN(-ENOMEM);
-
-        OBD_ALLOC(where, sizeof(*where) * oa_bufs);
-        if (!where)
-                GOTO(out_sinfo, rc = -ENOMEM);
-
-        if (oa) {
-                OBD_ALLOC(obdos, sizeof(*obdos) * stripe_count);
-                if (!obdos)
-                        GOTO(out_where, rc = -ENOMEM);
-        }
-
-        OBD_ALLOC(ioarr, sizeof(*ioarr) * oa_bufs);
-        if (!ioarr)
-                GOTO(out_obdos, rc = -ENOMEM);
-
-        for (i = 0; i < oa_bufs; i++) {
-                where[i] = lov_stripe_number(lsm, pga[i].disk_offset);
-                stripeinfo[where[i]].bufct++;
-        }
-
-        for (i = 0, loi = lsm->lsm_oinfo, si_last = si = stripeinfo;
-             i < stripe_count; i++, loi++, si_last = si, si++) {
-                if (i > 0)
-                        si->index = si_last->index + si_last->bufct;
-                si->lsm.lsm_object_id = loi->loi_id;
-                si->ost_idx = loi->loi_ost_idx;
-
-                if (oa) {
-                        memcpy(&obdos[i], oa, sizeof(*obdos));
-                        obdos[i].o_id = si->lsm.lsm_object_id;
-                }
-        }
-
-        for (i = 0; i < oa_bufs; i++) {
-                int which = where[i];
-                int shift;
-
-                shift = stripeinfo[which].index + stripeinfo[which].subcount;
-                LASSERT(shift < oa_bufs);
-                ioarr[shift] = pga[i];
-                lov_stripe_offset(lsm, pga[i].disk_offset, which,
-                                  &ioarr[shift].disk_offset);
-                stripeinfo[which].subcount++;
-        }
-
-        for (i = 0, si = stripeinfo; i < stripe_count; i++, si++) {
-                int shift = si->index;
-
-                if (si->bufct == 0)
-                        continue;
-
-                if (lov->tgts[si->ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", si->ost_idx);
-                        GOTO(out_ioarr, rc = -EIO);
-                }
-
-                LASSERT(shift < oa_bufs);
+        rc = lov_prep_brw_set(exp, oa, lsm, oa_bufs, pga, oti, &lovset);
+        if (rc)
+                RETURN(rc);
 
-                rc = obd_brw_async(cmd, lov->tgts[si->ost_idx].ltd_exp,
-                                   &obdos[i], &si->lsm, si->bufct,
-                                   &ioarr[shift], set, oti);
+        list_for_each (pos, &lovset->set_list) {
+                struct obd_export *sub_exp;
+                struct brw_page *sub_pga;
+                req = list_entry(pos, struct lov_request, rq_link);
+                
+                sub_exp = lov->tgts[req->rq_idx].ltd_exp;
+                sub_pga = lovset->set_pga + req->rq_pgaidx;
+                rc = obd_brw_async(cmd, sub_exp, req->rq_oa, req->rq_md,
+                                   req->rq_oabufs, sub_pga, set, oti);
                 if (rc)
-                        GOTO(out_ioarr, rc);
+                        GOTO(out, rc);
+                lov_update_common_set(lovset, req, rc);
         }
         LASSERT(rc == 0);
         LASSERT(set->set_interpret == NULL);
         set->set_interpret = (set_interpreter_func)lov_brw_interpret;
-        LASSERT(sizeof(set->set_args) >= sizeof(struct lov_brw_async_args));
-        aa = (struct lov_brw_async_args *)&set->set_args;
-        aa->aa_lsm = lsm;
-        aa->aa_obdos = obdos;
-        aa->aa_oa = oa;
-        aa->aa_ioarr = ioarr;
-        aa->aa_oa_bufs = oa_bufs;
-
-        /* Don't free ioarr or obdos - that's done in lov_brw_interpret */
-        GOTO(out_where, rc);
-
- out_ioarr:
-        OBD_FREE(ioarr, sizeof(*ioarr) * oa_bufs);
- out_obdos:
-        OBD_FREE(obdos, stripe_count * sizeof(*obdos));
- out_where:
-        OBD_FREE(where, sizeof(*where) * oa_bufs);
- out_sinfo:
-        OBD_FREE(stripeinfo, stripe_count * sizeof(*stripeinfo));
-        return rc;
+        set->set_arg = (void *)lovset;
+        
+        RETURN(rc);
+out:
+        lov_fini_brw_set(lovset);
+        RETURN(rc);
 }
 
-struct lov_async_page *lap_from_cookie(void *cookie)
-{
-        struct lov_async_page *lap = cookie;
-        if (lap->lap_magic != LAP_MAGIC)
-                return ERR_PTR(-EINVAL);
-        return lap;
-};
-
 static int lov_ap_make_ready(void *data, int cmd)
 {
-        struct lov_async_page *lap = lap_from_cookie(data);
-        /* XXX should these assert? */
-        if (IS_ERR(lap))
-                return -EINVAL;
+        struct lov_async_page *lap = LAP_FROM_COOKIE(data);
 
         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
 }
 static int lov_ap_refresh_count(void *data, int cmd)
 {
-        struct lov_async_page *lap = lap_from_cookie(data);
-        if (IS_ERR(lap))
-                return -EINVAL;
+        struct lov_async_page *lap = LAP_FROM_COOKIE(data);
 
         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
                                                      cmd);
 }
 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
 {
-        struct lov_async_page *lap = lap_from_cookie(data);
-        /* XXX should these assert? */
-        if (IS_ERR(lap))
-                return;
+        struct lov_async_page *lap = LAP_FROM_COOKIE(data);
 
         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
         /* XXX woah, shouldn't we be altering more here?  size? */
@@ -2157,9 +1339,7 @@ static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
 
 static void lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
 {
-        struct lov_async_page *lap = lap_from_cookie(data);
-        if (IS_ERR(lap))
-                return;
+        struct lov_async_page *lap = LAP_FROM_COOKIE(data);
 
         /* in a raid1 regime this would down a count of many ios
          * in flight, onl calling the caller_ops completion when all
@@ -2245,7 +1425,7 @@ static int lov_queue_async_io(struct obd_export *exp,
                               struct lov_stripe_md *lsm,
                               struct lov_oinfo *loi, void *cookie,
                               int cmd, obd_off off, int count,
-                              obd_flag brw_flags, obd_flag async_flags)
+                              obd_flags brw_flags, obd_flags async_flags)
 {
         struct lov_obd *lov = &exp->exp_obd->u.lov;
         struct lov_async_page *lap;
@@ -2256,9 +1436,7 @@ static int lov_queue_async_io(struct obd_export *exp,
         if (lsm_bad_magic(lsm))
                 RETURN(-EINVAL);
 
-        lap = lap_from_cookie(cookie);
-        if (IS_ERR(lap))
-                RETURN(PTR_ERR(lap));
+        lap = LAP_FROM_COOKIE(cookie);
 
         loi = &lsm->lsm_oinfo[lap->lap_stripe];
 
@@ -2271,7 +1449,7 @@ static int lov_queue_async_io(struct obd_export *exp,
 static int lov_set_async_flags(struct obd_export *exp,
                                struct lov_stripe_md *lsm,
                                struct lov_oinfo *loi, void *cookie,
-                               obd_flag async_flags)
+                               obd_flags async_flags)
 {
         struct lov_obd *lov = &exp->exp_obd->u.lov;
         struct lov_async_page *lap;
@@ -2282,9 +1460,7 @@ static int lov_set_async_flags(struct obd_export *exp,
         if (lsm_bad_magic(lsm))
                 RETURN(-EINVAL);
 
-        lap = lap_from_cookie(cookie);
-        if (IS_ERR(lap))
-                RETURN(PTR_ERR(lap));
+        lap = LAP_FROM_COOKIE(cookie);
 
         loi = &lsm->lsm_oinfo[lap->lap_stripe];
 
@@ -2298,7 +1474,7 @@ static int lov_queue_group_io(struct obd_export *exp,
                               struct lov_oinfo *loi,
                               struct obd_io_group *oig, void *cookie,
                               int cmd, obd_off off, int count,
-                              obd_flag brw_flags, obd_flag async_flags)
+                              obd_flags brw_flags, obd_flags async_flags)
 {
         struct lov_obd *lov = &exp->exp_obd->u.lov;
         struct lov_async_page *lap;
@@ -2309,9 +1485,7 @@ static int lov_queue_group_io(struct obd_export *exp,
         if (lsm_bad_magic(lsm))
                 RETURN(-EINVAL);
 
-        lap = lap_from_cookie(cookie);
-        if (IS_ERR(lap))
-                RETURN(PTR_ERR(lap));
+        lap = LAP_FROM_COOKIE(cookie);
 
         loi = &lsm->lsm_oinfo[lap->lap_stripe];
 
@@ -2365,9 +1539,7 @@ static int lov_teardown_async_page(struct obd_export *exp,
         if (lsm_bad_magic(lsm))
                 RETURN(-EINVAL);
 
-        lap = lap_from_cookie(cookie);
-        if (IS_ERR(lap))
-                RETURN(PTR_ERR(lap));
+        lap = LAP_FROM_COOKIE(cookie);
 
         loi = &lsm->lsm_oinfo[lap->lap_stripe];
 
@@ -2388,14 +1560,13 @@ static int lov_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
                        void *data,__u32 lvb_len, void *lvb_swabber,
                        struct lustre_handle *lockh)
 {
-        struct lov_lock_handles *lov_lockh = NULL;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
         struct lustre_handle *lov_lockhp;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        char submd_buf[sizeof(struct lov_stripe_md) + sizeof(struct lov_oinfo)];
-        struct lov_stripe_md *submd = (void *)submd_buf;
         ldlm_error_t rc;
-        int i, save_flags = *flags;
+        int save_flags = *flags;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
@@ -2407,145 +1578,44 @@ static int lov_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
-        if (lsm->lsm_stripe_count > 1) {
-                lov_lockh = lov_llh_new(lsm);
-                if (lov_lockh == NULL)
-                        RETURN(-ENOMEM);
-
-                lockh->cookie = lov_lockh->llh_handle.h_cookie;
-                lov_lockhp = lov_lockh->llh_handles;
-        } else {
-                lov_lockhp = lockh;
-        }
-
         lov = &exp->exp_obd->u.lov;
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++, lov_lockhp++) {
-                ldlm_policy_data_t sub_ext;
-                obd_off start, end;
-
-                if (!lov_stripe_intersects(lsm, i, policy->l_extent.start,
-                                           policy->l_extent.end, &start,
-                                           &end))
-                        continue;
-
-                sub_ext.l_extent.start = start;
-                sub_ext.l_extent.end = end;
-                sub_ext.l_extent.gid = policy->l_extent.gid;
+        rc = lov_prep_enqueue_set(exp, lsm, policy, mode, lockh, &set);
+        if (rc)
+                RETURN(rc);
 
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
+        list_for_each (pos, &set->set_list) {
+                ldlm_policy_data_t sub_policy;
+                req = list_entry(pos, struct lov_request, rq_link);
+                lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
+                LASSERT(lov_lockhp);
 
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd->lsm_object_id = loi->loi_id;
-                submd->lsm_object_gr = lsm->lsm_object_gr;
-                submd->lsm_stripe_count = 0;
-                submd->lsm_oinfo->loi_kms_valid = loi->loi_kms_valid;
-                submd->lsm_oinfo->loi_rss = loi->loi_rss;
-                submd->lsm_oinfo->loi_kms = loi->loi_kms;
-                submd->lsm_oinfo->loi_blocks = loi->loi_blocks;
-                loi->loi_mtime = submd->lsm_oinfo->loi_mtime;
-                /* XXX submd is not fully initialized here */
                 *flags = save_flags;
-                rc = obd_enqueue(lov->tgts[loi->loi_ost_idx].ltd_exp, submd,
-                                 type, &sub_ext, mode, flags, bl_cb, cp_cb,
-                                 gl_cb, data, lvb_len, lvb_swabber, lov_lockhp);
-
-                /* XXX FIXME: This unpleasantness doesn't belong here at *all*.
-                 * It belongs in the OSC, except that the OSC doesn't have
-                 * access to the real LOI -- it gets a copy, that we created
-                 * above, and that copy can be arbitrarily out of date.
-                 *
-                 * The LOV API is due for a serious rewriting anyways, and this
-                 * can be addressed then. */
-                if (rc == ELDLM_OK) {
-                        struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);
-                        __u64 tmp = submd->lsm_oinfo->loi_rss;
-
-                        LASSERT(lock != NULL);
-                        loi->loi_rss = tmp;
-                        loi->loi_blocks = submd->lsm_oinfo->loi_blocks;
-                        /* Extend KMS up to the end of this lock and no further
-                         * A lock on [x,y] means a KMS of up to y + 1 bytes! */
-                        if (tmp > lock->l_policy_data.l_extent.end)
-                                tmp = lock->l_policy_data.l_extent.end + 1;
-                        if (tmp >= loi->loi_kms) {
-                                CDEBUG(D_INODE, "lock acquired, setting rss="
-                                       LPU64", kms="LPU64"\n", loi->loi_rss,
-                                       tmp);
-                                loi->loi_kms = tmp;
-                                loi->loi_kms_valid = 1;
-                        } else {
-                                CDEBUG(D_INODE, "lock acquired, setting rss="
-                                       LPU64"; leaving kms="LPU64", end="LPU64
-                                       "\n", loi->loi_rss, loi->loi_kms,
-                                       lock->l_policy_data.l_extent.end);
-                        }
-                        ldlm_lock_allow_match(lock);
-                        LDLM_LOCK_PUT(lock);
-                } else if (rc == ELDLM_LOCK_ABORTED &&
-                           save_flags & LDLM_FL_HAS_INTENT) {
-                        memset(lov_lockhp, 0, sizeof(*lov_lockhp));
-                        loi->loi_rss = submd->lsm_oinfo->loi_rss;
-                        loi->loi_blocks = submd->lsm_oinfo->loi_blocks;
-                        CDEBUG(D_INODE, "glimpsed, setting rss="LPU64"; leaving"
-                               " kms="LPU64"\n", loi->loi_rss, loi->loi_kms);
-                } else {
-                        memset(lov_lockhp, 0, sizeof(*lov_lockhp));
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: enqueue objid "LPX64" subobj "
-                                       LPX64" on OST idx %d: rc = %d\n",
-                                       lsm->lsm_object_id, loi->loi_id,
-                                       loi->loi_ost_idx, rc);
-                                GOTO(out_locks, rc);
-                        }
-                }
-        }
-        if (lsm->lsm_stripe_count > 1)
-                lov_llh_put(lov_lockh);
-        RETURN(ELDLM_OK);
-
- out_locks:
-        while (loi--, lov_lockhp--, i-- > 0) {
-                struct lov_stripe_md submd;
-                int err;
-
-                if (lov_lockhp->cookie == 0)
-                        continue;
-
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd.lsm_object_id = loi->loi_id;
-                submd.lsm_object_gr = lsm->lsm_object_gr;
-                submd.lsm_stripe_count = 0;
-                err = obd_cancel(lov->tgts[loi->loi_ost_idx].ltd_exp, &submd,
-                                 mode, lov_lockhp);
-                if (err && lov->tgts[loi->loi_ost_idx].active) {
-                        CERROR("error: cancelling objid "LPX64" on OST "
-                               "idx %d after enqueue error: rc = %d\n",
-                               loi->loi_id, loi->loi_ost_idx, err);
-                }
+                sub_policy.l_extent.start = req->rq_extent.start;
+                sub_policy.l_extent.end = req->rq_extent.end;
+
+                rc = obd_enqueue(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
+                                 type, &sub_policy, mode, flags, bl_cb,
+                                 cp_cb, gl_cb, data, lvb_len, lvb_swabber,
+                                 lov_lockhp);
+                rc = lov_update_enqueue_set(set, req, rc, save_flags);
+                if (rc != ELDLM_OK)
+                        break;
         }
 
-        if (lsm->lsm_stripe_count > 1) {
-                lov_llh_destroy(lov_lockh);
-                lov_llh_put(lov_lockh);
-        }
-        return rc;
+        lov_fini_enqueue_set(set, mode);
+        RETURN(rc);
 }
 
 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
                      int *flags, void *data, struct lustre_handle *lockh)
 {
-        struct lov_lock_handles *lov_lockh = NULL;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
         struct lustre_handle *lov_lockhp;
-        struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        struct lov_stripe_md submd;
-        ldlm_error_t rc = 0;
-        int i;
+        int lov_flags, rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
@@ -2554,86 +1624,29 @@ static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
-        if (lsm->lsm_stripe_count > 1) {
-                lov_lockh = lov_llh_new(lsm);
-                if (lov_lockh == NULL)
-                        RETURN(-ENOMEM);
-
-                lockh->cookie = lov_lockh->llh_handle.h_cookie;
-                lov_lockhp = lov_lockh->llh_handles;
-        } else {
-                lov_lockhp = lockh;
-        }
-
         lov = &exp->exp_obd->u.lov;
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++, lov_lockhp++) {
-                ldlm_policy_data_t sub_ext;
-                obd_off start, end;
-                int lov_flags;
-
-                if (!lov_stripe_intersects(lsm, i, policy->l_extent.start,
-                                           policy->l_extent.end, &start, &end))
-                        continue;
-
-                sub_ext.l_extent.start = start;
-                sub_ext.l_extent.end = end;
+        rc = lov_prep_match_set(exp, lsm, policy, mode, lockh, &set);
+        if (rc)
+                RETURN(rc);
 
-                if (obd_uuid_empty(&lov->tgts[loi->loi_ost_idx].uuid)) {
-                        CDEBUG(D_HA, "lov idx %d deleted\n", loi->loi_ost_idx);
-                        continue;
-                }
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        rc = -EIO;
-                        break;
-                }
+        list_for_each (pos, &set->set_list) {
+                ldlm_policy_data_t sub_policy;
+                req = list_entry(pos, struct lov_request, rq_link);
+                lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
+                LASSERT(lov_lockhp);
 
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd.lsm_object_id = loi->loi_id;
-                submd.lsm_object_gr = lsm->lsm_object_gr;
-                submd.lsm_stripe_count = 0;
+                sub_policy.l_extent.start = req->rq_extent.start;
+                sub_policy.l_extent.end = req->rq_extent.end;
                 lov_flags = *flags;
-                /* XXX submd is not fully initialized here */
-                rc = obd_match(lov->tgts[loi->loi_ost_idx].ltd_exp, &submd,
-                               type, &sub_ext, mode, &lov_flags, data,
+
+                rc = obd_match(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
+                               type, &sub_policy, mode, &lov_flags, data,
                                lov_lockhp);
+                rc = lov_update_match_set(set, req, rc);
                 if (rc != 1)
                         break;
         }
-        if (rc == 1) {
-                if (lsm->lsm_stripe_count > 1) {
-                        if (*flags & LDLM_FL_TEST_LOCK)
-                                lov_llh_destroy(lov_lockh);
-                        lov_llh_put(lov_lockh);
-                }
-                RETURN(1);
-        }
-
-        while (loi--, lov_lockhp--, i-- > 0) {
-                struct lov_stripe_md submd;
-                int err;
-
-                if (lov_lockhp->cookie == 0)
-                        continue;
-
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd.lsm_object_id = loi->loi_id;
-                submd.lsm_object_gr = lsm->lsm_object_gr;
-                submd.lsm_stripe_count = 0;
-                err = obd_cancel(lov->tgts[loi->loi_ost_idx].ltd_exp, &submd,
-                                 mode, lov_lockhp);
-                if (err && lov->tgts[loi->loi_ost_idx].active) {
-                        CERROR("error: cancelling objid "LPX64" on OST "
-                               "idx %d after match failure: rc = %d\n",
-                               loi->loi_id, loi->loi_ost_idx, err);
-                }
-        }
-
-        if (lsm->lsm_stripe_count > 1) {
-                lov_llh_destroy(lov_lockh);
-                lov_llh_put(lov_lockh);
-        }
+        lov_fini_match_set(set, mode, *flags);
         RETURN(rc);
 }
 
@@ -2674,11 +1687,12 @@ static int lov_change_cbdata(struct obd_export *exp,
 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
                       __u32 mode, struct lustre_handle *lockh)
 {
-        struct lov_lock_handles *lov_lockh = NULL;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
         struct lustre_handle *lov_lockhp;
-        struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i;
+        int err = 0, rc = 0;
         ENTRY;
 
         if (lsm_bad_magic(lsm))
@@ -2690,57 +1704,34 @@ static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
         LASSERT(lsm->lsm_object_gr > 0);
 
         LASSERT(lockh);
-        if (lsm->lsm_stripe_count > 1) {
-                lov_lockh = lov_handle2llh(lockh);
-                if (!lov_lockh) {
-                        CERROR("LOV: invalid lov lock handle %p\n", lockh);
-                        RETURN(-EINVAL);
-                }
-
-                lov_lockhp = lov_lockh->llh_handles;
-        } else {
-                lov_lockhp = lockh;
-        }
-
         lov = &exp->exp_obd->u.lov;
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++, lov_lockhp++) {
-                struct lov_stripe_md submd;
-                int err;
+        rc = lov_prep_cancel_set(exp, lsm, mode, lockh, &set);
+        if (rc)
+                RETURN(rc);
 
-                if (lov_lockhp->cookie == 0) {
-                        CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
-                               loi->loi_ost_idx, loi->loi_id);
-                        continue;
-                }
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
 
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd.lsm_object_id = loi->loi_id;
-                submd.lsm_object_gr = lsm->lsm_object_gr;
-                submd.lsm_stripe_count = 0;
-                err = obd_cancel(lov->tgts[loi->loi_ost_idx].ltd_exp, &submd,
-                                 mode, lov_lockhp);
-                if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: cancel objid "LPX64" subobj "
-                                       LPX64" on OST idx %d: rc = %d\n",
-                                       lsm->lsm_object_id,
-                                       loi->loi_id, loi->loi_ost_idx, err);
-                                if (!rc)
-                                        rc = err;
-                        }
+                rc = obd_cancel(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
+                                mode, lov_lockhp);
+                rc = lov_update_common_set(set, req, rc);
+                if (rc) {
+                        CERROR("error: cancel objid "LPX64" subobj "
+                               LPX64" on OST idx %d: rc = %d\n",
+                               lsm->lsm_object_id,
+                               req->rq_md->lsm_object_id, req->rq_idx, rc);
+                        err = rc;
                 }
         }
-
-        if (lsm->lsm_stripe_count > 1)
-                lov_llh_destroy(lov_lockh);
-        if (lov_lockh != NULL)
-                lov_llh_put(lov_lockh);
-        RETURN(rc);
+        lov_fini_cancel_set(set);
+        RETURN(err);
 }
 
 static int lov_cancel_unused(struct obd_export *exp,
-                             struct lov_stripe_md *lsm, int flags, void *opaque)
+                             struct lov_stripe_md *lsm, 
+                            int flags, void *opaque)
 {
         struct lov_obd *lov;
         struct lov_oinfo *loi;
@@ -2750,8 +1741,8 @@ static int lov_cancel_unused(struct obd_export *exp,
         lov = &exp->exp_obd->u.lov;
         if (lsm == NULL) {
                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
-                        int err = obd_cancel_unused(lov->tgts[i].ltd_exp, NULL,
-                                                    flags, opaque);
+                        int err = obd_cancel_unused(lov->tgts[i].ltd_exp,
+                                                    NULL, flags, opaque);
                         if (!rc)
                                 rc = err;
                 }
@@ -2910,7 +1901,7 @@ static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 tgtdesc = lov->tgts;
                 /* the uuid will be empty for deleted OSTs */
                 for (i = 0; i < count; i++, uuidp++, genp++, tgtdesc++) {
-                        obd_str2uuid(uuidp, tgtdesc->uuid.uuid);
+                        obd_str2uuid(uuidp, (char *)tgtdesc->uuid.uuid);
                         *genp = tgtdesc->ltd_gen;
                 }
 
@@ -2945,8 +1936,8 @@ static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                                             len, karg, uarg);
                         if (err) {
                                 if (lov->tgts[i].active) {
-                                        CERROR("error: iocontrol OSC %s on OST"
-                                               "idx %d: cmd %x err = %d\n",
+                                        CERROR("error: iocontrol OSC %s on OST "
+                                               "idx %d cmd %x: err = %d\n",
                                                lov->tgts[i].uuid.uuid, i,
                                                cmd, err);
                                         if (!rc)
@@ -3024,13 +2015,15 @@ static int lov_get_info(struct obd_export *exp, __u32 keylen,
                                                 data->stripe_number);
                 RETURN(0);
         } else if (keylen >= strlen("last_id") && strcmp(key, "last_id") == 0) {
+                __u32 size = sizeof(obd_id);
                 obd_id *ids = val;
-                int rc, size = sizeof(obd_id);
+                int rc = 0;
+
                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
                         if (!lov->tgts[i].active)
                                 continue;
-                        rc = obd_get_info(lov->tgts[i].ltd_exp, keylen, key,
-                                          &size, &(ids[i]));
+                        rc = obd_get_info(lov->tgts[i].ltd_exp,
+                                          keylen, key, &size, &(ids[i]));
                         if (rc != 0)
                                 RETURN(rc);
                 }
@@ -3050,120 +2043,122 @@ static int lov_set_info(struct obd_export *exp, obd_count keylen,
 {
         struct obd_device *obddev = class_exp2obd(exp);
         struct lov_obd *lov = &obddev->u.lov;
-        int i, rc = 0;
+        int i, rc = 0, err;
         ENTRY;
 
 #define KEY_IS(str) \
         (keylen == strlen(str) && memcmp(key, str, keylen) == 0)
 
-        if (KEY_IS("next_id")) {
-                if (vallen != lov->desc.ld_tgt_count)
-                        RETURN(-EINVAL);
-                for (i = 0; i < lov->desc.ld_tgt_count; i++) {
-                        int er;
+        if (KEY_IS("async")) {
+                struct lov_desc *desc = &lov->desc;
+                struct lov_tgt_desc *tgts = lov->tgts;
 
-                        /* OST was deleted */
-                        if (obd_uuid_empty(&lov->tgts[i].uuid))
+                if (vallen != sizeof(int))
+                        RETURN(-EINVAL);
+                lov->async = *((int*) val);
+
+                for (i = 0; i < desc->ld_tgt_count; i++, tgts++) {
+                        struct obd_uuid *tgt_uuid = &tgts->uuid;
+                        struct obd_device *tgt_obd;
+
+                        tgt_obd = class_find_client_obd(tgt_uuid,
+                                                        LUSTRE_OSC_NAME,
+                                                        &obddev->obd_uuid);
+                        if (!tgt_obd) {
+                                CERROR("Target %s not attached\n",
+                                        tgt_uuid->uuid);
+                                if (!rc)
+                                        rc = -EINVAL;
                                 continue;
+                        }
 
-                        /* initialize all OSCs, even inactive ones */
-
-                        er = obd_set_info(lov->tgts[i].ltd_exp, keylen, key,
-                                          sizeof(obd_id), ((obd_id*)val) + i);
-                        if (!rc)
-                                rc = er;
+                        err = obd_set_info(tgt_obd->obd_self_export,
+                                           keylen, key, vallen, val);
+                        if (err) {
+                                CERROR("Failed to set async on target %s\n",
+                                        tgt_obd->obd_name);
+                                if (!rc)
+                                        rc = err;
+                        }
                 }
                 RETURN(rc);
         }
-        if (KEY_IS("growth_count")) {
-                if (vallen != sizeof(int))
-                        RETURN(-EINVAL);
-        } else if (KEY_IS("mds_conn")) {
+
+        if (KEY_IS("mds_conn")) {
                 if (vallen != sizeof(__u32))
                         RETURN(-EINVAL);
         } else if (KEY_IS("unlinked") || KEY_IS("unrecovery")) {
                 if (vallen != 0)
                         RETURN(-EINVAL);
+        } else if (KEY_IS("sec")) {
+                struct lov_tgt_desc *tgt;
+                struct obd_export *exp;
+                int rc = 0, err, i;
+
+                spin_lock(&lov->lov_lock);
+                for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count;
+                     i++, tgt++) {
+                        exp = tgt->ltd_exp;
+                        /* during setup time the connections to osc might
+                         * haven't been established.
+                         */
+                        if (exp == NULL) {
+                                struct obd_device *tgt_obd;
+
+                                tgt_obd = class_find_client_obd(&tgt->uuid,
+                                                                LUSTRE_OSC_NAME,
+                                                                &obddev->obd_uuid);
+                                if (!tgt_obd) {
+                                        CERROR("can't set security flavor, "
+                                               "device %s not attached?\n",
+                                                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(&lov->lov_lock);
+
+                RETURN(rc);
+        } else if (KEY_IS("flush_cred")) {
+                struct lov_tgt_desc *tgt;
+                int rc = 0, i;
+
+                for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count;
+                     i++, tgt++) {
+                        if (!tgt->ltd_exp)
+                                continue;
+                        rc = obd_set_info(tgt->ltd_exp,
+                                          keylen, key, vallen, val);
+                        if (rc)
+                                RETURN(rc);
+                }
+
+                RETURN(0);
         } else {
                 RETURN(-EINVAL);
         }
 
         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
-                int er;
-
-                /* OST was deleted */
-                if (obd_uuid_empty(&lov->tgts[i].uuid))
+                if (val && !obd_uuid_equals(val, &lov->tgts[i].uuid))
                         continue;
 
                 if (!val && !lov->tgts[i].active)
                         continue;
 
-                er = obd_set_info(lov->tgts[i].ltd_exp, keylen, key, vallen,
-                                  val);
+                err = obd_set_info(lov->tgts[i].ltd_exp,
+                                  keylen, key, vallen, val);
                 if (!rc)
-                        rc = er;
+                        rc = err;
         }
         RETURN(rc);
 #undef KEY_IS
-
-}
-
-/* Merge rss if @kms_only == 0
- *
- * Even when merging RSS, we will take the KMS value if it's larger.
- * This prevents getattr from stomping on dirty cached pages which
- * extend the file size. */
-__u64 lov_merge_size(struct lov_stripe_md *lsm, int kms_only)
-{
-        struct lov_oinfo *loi;
-        __u64 size = 0;
-        int i;
-
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++) {
-                obd_size lov_size, tmpsize;
-
-                tmpsize = loi->loi_kms;
-                if (kms_only == 0 && loi->loi_rss > tmpsize)
-                        tmpsize = loi->loi_rss;
-
-                lov_size = lov_stripe_size(lsm, tmpsize, i);
-                if (lov_size > size)
-                        size = lov_size;
-        }
-
-        return size;
-}
-EXPORT_SYMBOL(lov_merge_size);
-
-/* Merge blocks */
-__u64 lov_merge_blocks(struct lov_stripe_md *lsm)
-{
-        struct lov_oinfo *loi;
-        __u64 blocks = 0;
-        int i;
-
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++) {
-                blocks += loi->loi_blocks;
-        }
-        return blocks;
 }
-EXPORT_SYMBOL(lov_merge_blocks);
-
-__u64 lov_merge_mtime(struct lov_stripe_md *lsm, __u64 current_time)
-{
-        struct lov_oinfo *loi;
-        int i;
-
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++) {
-                if (loi->loi_mtime > current_time)
-                        current_time = loi->loi_mtime;
-        }
-        return current_time;
-}
-EXPORT_SYMBOL(lov_merge_mtime);
 
 #if 0
 struct lov_multi_wait {
@@ -3255,27 +2250,6 @@ int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
 }
 #endif
 
-void lov_increase_kms(struct obd_export *exp, struct lov_stripe_md *lsm,
-                      obd_off size)
-{
-        struct lov_oinfo *loi;
-        int stripe = 0;
-        __u64 kms;
-        ENTRY;
-
-        if (size > 0)
-                stripe = lov_stripe_number(lsm, size - 1);
-        kms = lov_size_to_stripe(lsm, size, stripe);
-        loi = &(lsm->lsm_oinfo[stripe]);
-
-        CDEBUG(D_INODE, "stripe %d KMS %sincreasing "LPU64"->"LPU64"\n",
-               stripe, kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms);
-        if (kms > loi->loi_kms)
-                loi->loi_kms = kms;
-        EXIT;
-}
-EXPORT_SYMBOL(lov_increase_kms);
-
 struct obd_ops lov_obd_ops = {
         .o_owner               = THIS_MODULE,
         .o_attach              = lov_attach,
@@ -3302,6 +2276,7 @@ struct obd_ops lov_obd_ops = {
         .o_queue_group_io      = lov_queue_group_io,
         .o_trigger_group_io    = lov_trigger_group_io,
         .o_teardown_async_page = lov_teardown_async_page,
+        .o_adjust_kms          = lov_adjust_kms,
         .o_punch               = lov_punch,
         .o_sync                = lov_sync,
         .o_enqueue             = lov_enqueue,