Whamcloud - gitweb
LU-3086 build: fix 'uninitialized variables' errors
[fs/lustre-release.git] / lustre / osc / osc_request.c
index aa79f52..4092d93 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2012, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -58,6 +58,7 @@
 #include <lustre_log.h>
 #include <lustre_debug.h>
 #include <lustre_param.h>
+#include <lustre_fid.h>
 #include "osc_internal.h"
 #include "osc_cl_internal.h"
 
@@ -68,97 +69,95 @@ int osc_cleanup(struct obd_device *obd);
 
 /* Pack OSC object metadata for disk storage (LE byte order). */
 static int osc_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
-                      struct lov_stripe_md *lsm)
+                     struct lov_stripe_md *lsm)
 {
-        int lmm_size;
-        ENTRY;
+       int lmm_size;
+       ENTRY;
 
-        lmm_size = sizeof(**lmmp);
-        if (!lmmp)
-                RETURN(lmm_size);
+       lmm_size = sizeof(**lmmp);
+       if (lmmp == NULL)
+               RETURN(lmm_size);
 
-        if (*lmmp && !lsm) {
-                OBD_FREE(*lmmp, lmm_size);
-                *lmmp = NULL;
-                RETURN(0);
-        }
+       if (*lmmp != NULL && lsm == NULL) {
+               OBD_FREE(*lmmp, lmm_size);
+               *lmmp = NULL;
+               RETURN(0);
+       } else if (unlikely(lsm != NULL && ostid_id(&lsm->lsm_oi) == 0)) {
+               RETURN(-EBADF);
+       }
 
-        if (!*lmmp) {
-                OBD_ALLOC(*lmmp, lmm_size);
-                if (!*lmmp)
-                        RETURN(-ENOMEM);
-        }
+       if (*lmmp == NULL) {
+               OBD_ALLOC(*lmmp, lmm_size);
+               if (*lmmp == NULL)
+                       RETURN(-ENOMEM);
+       }
 
-        if (lsm) {
-                LASSERT(lsm->lsm_object_id);
-                LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
-                (*lmmp)->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
-                (*lmmp)->lmm_object_seq = cpu_to_le64(lsm->lsm_object_seq);
-        }
+       if (lsm)
+               ostid_cpu_to_le(&lsm->lsm_oi, &(*lmmp)->lmm_oi);
 
-        RETURN(lmm_size);
+       RETURN(lmm_size);
 }
 
 /* Unpack OSC object metadata from disk storage (LE byte order). */
 static int osc_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
-                        struct lov_mds_md *lmm, int lmm_bytes)
+                       struct lov_mds_md *lmm, int lmm_bytes)
 {
-        int lsm_size;
-        struct obd_import *imp = class_exp2cliimp(exp);
-        ENTRY;
+       int lsm_size;
+       struct obd_import *imp = class_exp2cliimp(exp);
+       ENTRY;
 
-        if (lmm != NULL) {
-                if (lmm_bytes < sizeof (*lmm)) {
-                        CERROR("lov_mds_md too small: %d, need %d\n",
-                               lmm_bytes, (int)sizeof(*lmm));
-                        RETURN(-EINVAL);
-                }
-                /* XXX LOV_MAGIC etc check? */
+       if (lmm != NULL) {
+               if (lmm_bytes < sizeof(*lmm)) {
+                       CERROR("%s: lov_mds_md too small: %d, need %d\n",
+                              exp->exp_obd->obd_name, lmm_bytes,
+                              (int)sizeof(*lmm));
+                       RETURN(-EINVAL);
+               }
+               /* XXX LOV_MAGIC etc check? */
 
-                if (lmm->lmm_object_id == 0) {
-                        CERROR("lov_mds_md: zero lmm_object_id\n");
-                        RETURN(-EINVAL);
-                }
-        }
+               if (unlikely(ostid_id(&lmm->lmm_oi) == 0)) {
+                       CERROR("%s: zero lmm_object_id: rc = %d\n",
+                              exp->exp_obd->obd_name, -EINVAL);
+                       RETURN(-EINVAL);
+               }
+       }
 
-        lsm_size = lov_stripe_md_size(1);
-        if (lsmp == NULL)
-                RETURN(lsm_size);
+       lsm_size = lov_stripe_md_size(1);
+       if (lsmp == NULL)
+               RETURN(lsm_size);
 
-        if (*lsmp != NULL && lmm == NULL) {
-                OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
-                OBD_FREE(*lsmp, lsm_size);
-                *lsmp = NULL;
-                RETURN(0);
-        }
+       if (*lsmp != NULL && lmm == NULL) {
+               OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
+               OBD_FREE(*lsmp, lsm_size);
+               *lsmp = NULL;
+               RETURN(0);
+       }
 
-        if (*lsmp == NULL) {
-                OBD_ALLOC(*lsmp, lsm_size);
-                if (*lsmp == NULL)
-                        RETURN(-ENOMEM);
-                OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
-                if ((*lsmp)->lsm_oinfo[0] == NULL) {
-                        OBD_FREE(*lsmp, lsm_size);
-                        RETURN(-ENOMEM);
-                }
-                loi_init((*lsmp)->lsm_oinfo[0]);
-        }
+       if (*lsmp == NULL) {
+               OBD_ALLOC(*lsmp, lsm_size);
+               if (unlikely(*lsmp == NULL))
+                       RETURN(-ENOMEM);
+               OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
+               if (unlikely((*lsmp)->lsm_oinfo[0] == NULL)) {
+                       OBD_FREE(*lsmp, lsm_size);
+                       RETURN(-ENOMEM);
+               }
+               loi_init((*lsmp)->lsm_oinfo[0]);
+       } else if (unlikely(ostid_id(&(*lsmp)->lsm_oi) == 0)) {
+               RETURN(-EBADF);
+       }
 
-        if (lmm != NULL) {
-                /* XXX zero *lsmp? */
-                (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
-                (*lsmp)->lsm_object_seq = le64_to_cpu (lmm->lmm_object_seq);
-                LASSERT((*lsmp)->lsm_object_id);
-                LASSERT_SEQ_IS_MDT((*lsmp)->lsm_object_seq);
-        }
+       if (lmm != NULL)
+               /* XXX zero *lsmp? */
+               ostid_le_to_cpu(&lmm->lmm_oi, &(*lsmp)->lsm_oi);
 
-        if (imp != NULL &&
-            (imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_MAXBYTES))
-                (*lsmp)->lsm_maxbytes = imp->imp_connect_data.ocd_maxbytes;
-        else
-                (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
+       if (imp != NULL &&
+           (imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_MAXBYTES))
+               (*lsmp)->lsm_maxbytes = imp->imp_connect_data.ocd_maxbytes;
+       else
+               (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
 
-        RETURN(lsm_size);
+       RETURN(lsm_size);
 }
 
 static inline void osc_pack_capa(struct ptlrpc_request *req,
@@ -215,9 +214,9 @@ static int osc_getattr_interpret(const struct lu_env *env,
                 CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
                 lustre_get_wire_obdo(aa->aa_oi->oi_oa, &body->oa);
 
-                /* This should really be sent by the OST */
-                aa->aa_oi->oi_oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
-                aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
+               /* This should really be sent by the OST */
+               aa->aa_oi->oi_oa->o_blksize = DT_MAX_BRW_SIZE;
+               aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
         } else {
                 CDEBUG(D_INFO, "can't unpack ost_body\n");
                 rc = -EPROTO;
@@ -294,9 +293,8 @@ static int osc_getattr(const struct lu_env *env, struct obd_export *exp,
         CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
         lustre_get_wire_obdo(oinfo->oi_oa, &body->oa);
 
-        /* This should really be sent by the OST */
-        oinfo->oi_oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
-        oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
+       oinfo->oi_oa->o_blksize = cli_brw_size(exp->exp_obd);
+       oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
 
         EXIT;
  out:
@@ -477,17 +475,15 @@ int osc_real_create(struct obd_export *exp, struct obdo *oa,
 
         lustre_get_wire_obdo(oa, &body->oa);
 
-        /* This should really be sent by the OST */
-        oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
-        oa->o_valid |= OBD_MD_FLBLKSZ;
+       oa->o_blksize = cli_brw_size(exp->exp_obd);
+       oa->o_valid |= OBD_MD_FLBLKSZ;
 
-        /* XXX LOV STACKING: the lsm that is passed to us from LOV does not
-         * have valid lsm_oinfo data structs, so don't go touching that.
-         * This needs to be fixed in a big way.
-         */
-        lsm->lsm_object_id = oa->o_id;
-        lsm->lsm_object_seq = oa->o_seq;
-        *ea = lsm;
+       /* XXX LOV STACKING: the lsm that is passed to us from LOV does not
+        * have valid lsm_oinfo data structs, so don't go touching that.
+        * This needs to be fixed in a big way.
+        */
+       lsm->lsm_oi = oa->o_oi;
+       *ea = lsm;
 
         if (oti != NULL) {
                 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
@@ -662,10 +658,19 @@ static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
         int count;
         ENTRY;
 
-        osc_build_res_name(oa->o_id, oa->o_seq, &res_id);
-        res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
-        if (res == NULL)
-                RETURN(0);
+       /* Return, i.e. cancel nothing, only if ELC is supported (flag in
+        * export) but disabled through procfs (flag in NS).
+        *
+        * This distinguishes from a case when ELC is not supported originally,
+        * when we still want to cancel locks in advance and just cancel them
+        * locally, without sending any RPC. */
+       if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
+               RETURN(0);
+
+       ostid_build_res_name(&oa->o_oi, &res_id);
+       res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
+       if (res == NULL)
+               RETURN(0);
 
         LDLM_RESOURCE_ADDREF(res);
         count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
@@ -704,6 +709,31 @@ static int osc_can_send_destroy(struct client_obd *cli)
         return 0;
 }
 
+int osc_create(const struct lu_env *env, struct obd_export *exp,
+              struct obdo *oa, struct lov_stripe_md **ea,
+              struct obd_trans_info *oti)
+{
+       int rc = 0;
+       ENTRY;
+
+       LASSERT(oa);
+       LASSERT(ea);
+       LASSERT(oa->o_valid & OBD_MD_FLGROUP);
+
+       if ((oa->o_valid & OBD_MD_FLFLAGS) &&
+           oa->o_flags == OBD_FL_RECREATE_OBJS) {
+               RETURN(osc_real_create(exp, oa, ea, oti));
+       }
+
+       if (!fid_seq_is_mdt(ostid_seq(&oa->o_oi)))
+               RETURN(osc_real_create(exp, oa, ea, oti));
+
+       /* we should not get here anymore */
+       LBUG();
+
+       RETURN(rc);
+}
+
 /* Destroy requests can be async always on the client, and we don't even really
  * care about the return code since the client cannot do anything at all about
  * a destroy failure.
@@ -794,28 +824,33 @@ static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
         oa->o_valid |= bits;
         client_obd_list_lock(&cli->cl_loi_list_lock);
         oa->o_dirty = cli->cl_dirty;
-        if (cli->cl_dirty - cli->cl_dirty_transit > cli->cl_dirty_max) {
-                CERROR("dirty %lu - %lu > dirty_max %lu\n",
-                       cli->cl_dirty, cli->cl_dirty_transit, cli->cl_dirty_max);
-                oa->o_undirty = 0;
-        } else if (cfs_atomic_read(&obd_dirty_pages) -
-                   cfs_atomic_read(&obd_dirty_transit_pages) >
-                   obd_max_dirty_pages + 1){
-                /* The cfs_atomic_read() allowing the cfs_atomic_inc() are
-                 * not covered by a lock thus they may safely race and trip
-                 * this CERROR() unless we add in a small fudge factor (+1). */
-                CERROR("dirty %d - %d > system dirty_max %d\n",
-                       cfs_atomic_read(&obd_dirty_pages),
-                       cfs_atomic_read(&obd_dirty_transit_pages),
-                       obd_max_dirty_pages);
-                oa->o_undirty = 0;
-        } else if (cli->cl_dirty_max - cli->cl_dirty > 0x7fffffff) {
-                CERROR("dirty %lu - dirty_max %lu too big???\n",
-                       cli->cl_dirty, cli->cl_dirty_max);
-                oa->o_undirty = 0;
-        } else {
-                long max_in_flight = (cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT)*
-                                (cli->cl_max_rpcs_in_flight + 1);
+       if (unlikely(cli->cl_dirty - cli->cl_dirty_transit >
+                    cli->cl_dirty_max)) {
+               CERROR("dirty %lu - %lu > dirty_max %lu\n",
+                      cli->cl_dirty, cli->cl_dirty_transit, cli->cl_dirty_max);
+               oa->o_undirty = 0;
+       } else if (unlikely(cfs_atomic_read(&obd_unstable_pages) +
+                           cfs_atomic_read(&obd_dirty_pages) -
+                           cfs_atomic_read(&obd_dirty_transit_pages) >
+                           (long)(obd_max_dirty_pages + 1))) {
+               /* The cfs_atomic_read() allowing the cfs_atomic_inc() are
+                * not covered by a lock thus they may safely race and trip
+                * this CERROR() unless we add in a small fudge factor (+1). */
+               CERROR("%s: dirty %d + %d - %d > system dirty_max %d\n",
+                      cli->cl_import->imp_obd->obd_name,
+                      cfs_atomic_read(&obd_unstable_pages),
+                      cfs_atomic_read(&obd_dirty_pages),
+                      cfs_atomic_read(&obd_dirty_transit_pages),
+                      obd_max_dirty_pages);
+               oa->o_undirty = 0;
+       } else if (unlikely(cli->cl_dirty_max - cli->cl_dirty > 0x7fffffff)) {
+               CERROR("dirty %lu - dirty_max %lu too big???\n",
+                      cli->cl_dirty, cli->cl_dirty_max);
+               oa->o_undirty = 0;
+       } else {
+               long max_in_flight = (cli->cl_max_pages_per_rpc <<
+                                     CFS_PAGE_SHIFT)*
+                                    (cli->cl_max_rpcs_in_flight + 1);
                 oa->o_undirty = max(cli->cl_dirty_max, max_in_flight);
         }
        oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
@@ -895,45 +930,45 @@ static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
  * needed, and avoids shrinking the grant piecemeal. */
 static int osc_shrink_grant(struct client_obd *cli)
 {
-        long target = (cli->cl_max_rpcs_in_flight + 1) *
-                      cli->cl_max_pages_per_rpc;
+       __u64 target_bytes = (cli->cl_max_rpcs_in_flight + 1) *
+                            (cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT);
 
-        client_obd_list_lock(&cli->cl_loi_list_lock);
-        if (cli->cl_avail_grant <= target)
-                target = cli->cl_max_pages_per_rpc;
-        client_obd_list_unlock(&cli->cl_loi_list_lock);
+       client_obd_list_lock(&cli->cl_loi_list_lock);
+       if (cli->cl_avail_grant <= target_bytes)
+               target_bytes = cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT;
+       client_obd_list_unlock(&cli->cl_loi_list_lock);
 
-        return osc_shrink_grant_to_target(cli, target);
+       return osc_shrink_grant_to_target(cli, target_bytes);
 }
 
-int osc_shrink_grant_to_target(struct client_obd *cli, long target)
+int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
 {
-        int    rc = 0;
-        struct ost_body     *body;
-        ENTRY;
-
-        client_obd_list_lock(&cli->cl_loi_list_lock);
-        /* Don't shrink if we are already above or below the desired limit
-         * We don't want to shrink below a single RPC, as that will negatively
-         * impact block allocation and long-term performance. */
-        if (target < cli->cl_max_pages_per_rpc)
-                target = cli->cl_max_pages_per_rpc;
+       int                     rc = 0;
+       struct ost_body        *body;
+       ENTRY;
 
-        if (target >= cli->cl_avail_grant) {
-                client_obd_list_unlock(&cli->cl_loi_list_lock);
-                RETURN(0);
-        }
-        client_obd_list_unlock(&cli->cl_loi_list_lock);
+       client_obd_list_lock(&cli->cl_loi_list_lock);
+       /* Don't shrink if we are already above or below the desired limit
+        * We don't want to shrink below a single RPC, as that will negatively
+        * impact block allocation and long-term performance. */
+       if (target_bytes < cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT)
+               target_bytes = cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT;
+
+       if (target_bytes >= cli->cl_avail_grant) {
+               client_obd_list_unlock(&cli->cl_loi_list_lock);
+               RETURN(0);
+       }
+       client_obd_list_unlock(&cli->cl_loi_list_lock);
 
-        OBD_ALLOC_PTR(body);
-        if (!body)
-                RETURN(-ENOMEM);
+       OBD_ALLOC_PTR(body);
+       if (!body)
+               RETURN(-ENOMEM);
 
-        osc_announce_cached(cli, &body->oa, 0);
+       osc_announce_cached(cli, &body->oa, 0);
 
-        client_obd_list_lock(&cli->cl_loi_list_lock);
-        body->oa.o_grant = cli->cl_avail_grant - target;
-        cli->cl_avail_grant = target;
+       client_obd_list_lock(&cli->cl_loi_list_lock);
+       body->oa.o_grant = cli->cl_avail_grant - target_bytes;
+       cli->cl_avail_grant = target_bytes;
         client_obd_list_unlock(&cli->cl_loi_list_lock);
         if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
                 body->oa.o_valid |= OBD_MD_FLFLAGS;
@@ -951,7 +986,6 @@ int osc_shrink_grant_to_target(struct client_obd *cli, long target)
         RETURN(rc);
 }
 
-#define GRANT_SHRINK_LIMIT PTLRPC_MAX_BRW_SIZE
 static int osc_should_shrink_grant(struct client_obd *client)
 {
         cfs_time_t time = cfs_time_current();
@@ -961,13 +995,18 @@ static int osc_should_shrink_grant(struct client_obd *client)
              OBD_CONNECT_GRANT_SHRINK) == 0)
                 return 0;
 
-        if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
-                if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
-                    client->cl_avail_grant > GRANT_SHRINK_LIMIT)
-                        return 1;
-                else
-                        osc_update_next_shrink(client);
-        }
+       if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
+               /* Get the current RPC size directly, instead of going via:
+                * cli_brw_size(obd->u.cli.cl_import->imp_obd->obd_self_export)
+                * Keep comment here so that it can be found by searching. */
+               int brw_size = client->cl_max_pages_per_rpc << CFS_PAGE_SHIFT;
+
+               if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
+                   client->cl_avail_grant > brw_size)
+                       return 1;
+               else
+                       osc_update_next_shrink(client);
+       }
         return 0;
 }
 
@@ -1025,12 +1064,12 @@ static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
                 cli->cl_avail_grant = ocd->ocd_grant - cli->cl_dirty;
 
         if (cli->cl_avail_grant < 0) {
-                CWARN("%s: available grant < 0, the OSS is probably not running"
-                      " with patch from bug20278 (%ld) \n",
-                      cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant);
-                /* workaround for 1.6 servers which do not have
-                 * the patch from bug20278 */
-                cli->cl_avail_grant = ocd->ocd_grant;
+               CWARN("%s: available grant < 0: avail/ocd/dirty %ld/%u/%ld\n",
+                     cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant,
+                     ocd->ocd_grant, cli->cl_dirty);
+               /* workaround for servers which do not have the patch from
+                * LU-2679 */
+               cli->cl_avail_grant = ocd->ocd_grant;
         }
 
        /* determine the appropriate chunk size used by osc_extent. */
@@ -1175,8 +1214,8 @@ static obd_count osc_checksum_bulk(int nob, obd_count pg_count,
                cfs_crypto_hash_update_page(hdesc, pga[i]->pg,
                                  pga[i]->off & ~CFS_PAGE_MASK,
                                  count);
-               LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d checksum %x\n",
-                              (int)(pga[i]->off & ~CFS_PAGE_MASK), cksum);
+               LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d\n",
+                              (int)(pga[i]->off & ~CFS_PAGE_MASK));
 
                nob -= pga[i]->count;
                pg_count--;
@@ -1255,12 +1294,10 @@ static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa,
         * retry logic */
        req->rq_no_retry_einprogress = 1;
 
-        if (opc == OST_WRITE)
-                desc = ptlrpc_prep_bulk_imp(req, page_count,
-                                            BULK_GET_SOURCE, OST_BULK_PORTAL);
-        else
-                desc = ptlrpc_prep_bulk_imp(req, page_count,
-                                            BULK_PUT_SINK, OST_BULK_PORTAL);
+       desc = ptlrpc_prep_bulk_imp(req, page_count,
+               cli->cl_import->imp_connect_data.ocd_brw_size >> LNET_MTU_BITS,
+               opc == OST_WRITE ? BULK_GET_SOURCE : BULK_PUT_SINK,
+               OST_BULK_PORTAL);
 
         if (desc == NULL)
                 GOTO(out, rc = -ENOMEM);
@@ -1273,11 +1310,17 @@ static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa,
 
         lustre_set_wire_obdo(&body->oa, oa);
 
-        obdo_to_ioobj(oa, ioobj);
-        ioobj->ioo_bufcnt = niocount;
-        osc_pack_capa(req, body, ocapa);
-        LASSERT (page_count > 0);
-        pg_prev = pga[0];
+       obdo_to_ioobj(oa, ioobj);
+       ioobj->ioo_bufcnt = niocount;
+       /* The high bits of ioo_max_brw tells server _maximum_ number of bulks
+        * that might be send for this request.  The actual number is decided
+        * when the RPC is finally sent in ptlrpc_register_bulk(). It sends
+        * "max - 1" for old client compatibility sending "0", and also so the
+        * the actual maximum is a power-of-two number, not one less. LU-1431 */
+       ioobj_max_brw_set(ioobj, desc->bd_md_max_brw);
+       osc_pack_capa(req, body, ocapa);
+       LASSERT(page_count > 0);
+       pg_prev = pga[0];
         for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
                 struct brw_page *pg = pga[i];
                 int poff = pg->off & ~CFS_PAGE_MASK;
@@ -1306,7 +1349,7 @@ static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa,
                 LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
                         (pg->flag & OBD_BRW_SRVLOCK));
 
-                ptlrpc_prep_bulk_page(desc, pg->pg, poff, pg->count);
+               ptlrpc_prep_bulk_page_pin(desc, pg->pg, poff, pg->count);
                 requested_nob += pg->count;
 
                 if (i > 0 && can_merge_pages(pg_prev, pg)) {
@@ -1432,20 +1475,18 @@ static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer,
                 msg = "changed in transit AND doesn't match the original - "
                       "likely false positive due to mmap IO (bug 11742)";
 
-        LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID
-                           " object "LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
-                           msg, libcfs_nid2str(peer->nid),
-                           oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
-                           oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
-                           oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
-                           oa->o_id,
-                           oa->o_valid & OBD_MD_FLGROUP ? oa->o_seq : (__u64)0,
-                           pga[0]->off,
-                           pga[page_count-1]->off + pga[page_count-1]->count - 1);
-        CERROR("original client csum %x (type %x), server csum %x (type %x), "
-               "client csum now %x\n", client_cksum, client_cksum_type,
-               server_cksum, cksum_type, new_cksum);
-        return 1;
+       LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID
+                          " object "DOSTID" extent ["LPU64"-"LPU64"]\n",
+                          msg, libcfs_nid2str(peer->nid),
+                          oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
+                          oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
+                          oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
+                          POSTID(&oa->o_oi), pga[0]->off,
+                          pga[page_count-1]->off + pga[page_count-1]->count - 1);
+       CERROR("original client csum %x (type %x), server csum %x (type %x), "
+              "client csum now %x\n", client_cksum, client_cksum_type,
+              server_cksum, cksum_type, new_cksum);
+       return 1;
 }
 
 /* Note rc enters this function as number of bytes transferred */
@@ -1554,42 +1595,39 @@ static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
                         router = libcfs_nid2str(req->rq_bulk->bd_sender);
                 }
 
-                if (server_cksum == ~0 && rc > 0) {
-                        CERROR("Protocol error: server %s set the 'checksum' "
-                               "bit, but didn't send a checksum.  Not fatal, "
-                               "but please notify on http://bugs.whamcloud.com/\n",
-                               libcfs_nid2str(peer->nid));
-                } else if (server_cksum != client_cksum) {
-                        LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
-                                           "%s%s%s inode "DFID" object "
-                                           LPU64"/"LPU64" extent "
-                                           "["LPU64"-"LPU64"]\n",
-                                           req->rq_import->imp_obd->obd_name,
-                                           libcfs_nid2str(peer->nid),
-                                           via, router,
-                                           body->oa.o_valid & OBD_MD_FLFID ?
-                                                body->oa.o_parent_seq : (__u64)0,
-                                           body->oa.o_valid & OBD_MD_FLFID ?
-                                                body->oa.o_parent_oid : 0,
-                                           body->oa.o_valid & OBD_MD_FLFID ?
-                                                body->oa.o_parent_ver : 0,
-                                           body->oa.o_id,
-                                           body->oa.o_valid & OBD_MD_FLGROUP ?
-                                                body->oa.o_seq : (__u64)0,
-                                           aa->aa_ppga[0]->off,
-                                           aa->aa_ppga[aa->aa_page_count-1]->off +
-                                           aa->aa_ppga[aa->aa_page_count-1]->count -
-                                                                        1);
-                        CERROR("client %x, server %x, cksum_type %x\n",
-                               client_cksum, server_cksum, cksum_type);
-                        cksum_counter = 0;
-                        aa->aa_oa->o_cksum = client_cksum;
-                        rc = -EAGAIN;
-                } else {
-                        cksum_counter++;
-                        CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
-                        rc = 0;
-                }
+               if (server_cksum == ~0 && rc > 0) {
+                       CERROR("Protocol error: server %s set the 'checksum' "
+                              "bit, but didn't send a checksum.  Not fatal, "
+                              "but please notify on http://bugs.whamcloud.com/\n",
+                              libcfs_nid2str(peer->nid));
+               } else if (server_cksum != client_cksum) {
+                       LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
+                                          "%s%s%s inode "DFID" object "DOSTID
+                                          " extent ["LPU64"-"LPU64"]\n",
+                                          req->rq_import->imp_obd->obd_name,
+                                          libcfs_nid2str(peer->nid),
+                                          via, router,
+                                          body->oa.o_valid & OBD_MD_FLFID ?
+                                               body->oa.o_parent_seq : (__u64)0,
+                                          body->oa.o_valid & OBD_MD_FLFID ?
+                                               body->oa.o_parent_oid : 0,
+                                          body->oa.o_valid & OBD_MD_FLFID ?
+                                               body->oa.o_parent_ver : 0,
+                                          POSTID(&body->oa.o_oi),
+                                          aa->aa_ppga[0]->off,
+                                          aa->aa_ppga[aa->aa_page_count-1]->off +
+                                          aa->aa_ppga[aa->aa_page_count-1]->count -
+                                                                       1);
+                       CERROR("client %x, server %x, cksum_type %x\n",
+                              client_cksum, server_cksum, cksum_type);
+                       cksum_counter = 0;
+                       aa->aa_oa->o_cksum = client_cksum;
+                       rc = -EAGAIN;
+               } else {
+                       cksum_counter++;
+                       CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
+                       rc = 0;
+               }
         } else if (unlikely(client_cksum)) {
                 static int cksum_missed;
 
@@ -1653,17 +1691,17 @@ restart_bulk:
                 if (rc != -EINPROGRESS &&
                     !client_should_resend(resends, &exp->exp_obd->u.cli)) {
                         CERROR("%s: too many resend retries for object: "
-                               ""LPU64":"LPU64", rc = %d.\n",
-                               exp->exp_obd->obd_name, oa->o_id, oa->o_seq, rc);
-                        goto out;
-                }
-                if (generation !=
-                    exp->exp_obd->u.cli.cl_import->imp_generation) {
-                        CDEBUG(D_HA, "%s: resend cross eviction for object: "
-                               ""LPU64":"LPU64", rc = %d.\n",
-                               exp->exp_obd->obd_name, oa->o_id, oa->o_seq, rc);
+                               ""DOSTID", rc = %d.\n", exp->exp_obd->obd_name,
+                              POSTID(&oa->o_oi), rc);
                         goto out;
                 }
+               if (generation !=
+                   exp->exp_obd->u.cli.cl_import->imp_generation) {
+                       CDEBUG(D_HA, "%s: resend cross eviction for object: "
+                              ""DOSTID", rc = %d.\n", exp->exp_obd->obd_name,
+                              POSTID(&oa->o_oi), rc);
+                       goto out;
+               }
 
                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
                                        NULL);
@@ -1713,6 +1751,7 @@ static int osc_brw_redo_request(struct ptlrpc_request *request,
         aa->aa_resends++;
         new_req->rq_interpret_reply = request->rq_interpret_reply;
         new_req->rq_async_args = request->rq_async_args;
+       new_req->rq_commit_cb = request->rq_commit_cb;
        /* cap resend delay to the current request timeout, this is similar to
         * what ptlrpc does (see after_reply()) */
        if (aa->aa_resends > new_req->rq_timeout)
@@ -1728,6 +1767,7 @@ static int osc_brw_redo_request(struct ptlrpc_request *request,
        cfs_list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
        CFS_INIT_LIST_HEAD(&new_aa->aa_exts);
        cfs_list_splice_init(&aa->aa_exts, &new_aa->aa_exts);
+       new_aa->aa_resends = aa->aa_resends;
 
         cfs_list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
                 if (oap->oap_request) {
@@ -1914,28 +1954,28 @@ static int brw_interpret(const struct lu_env *env,
         CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
         /* When server return -EINPROGRESS, client should always retry
          * regardless of the number of times the bulk was resent already. */
-        if (osc_recoverable_error(rc)) {
-                if (req->rq_import_generation !=
-                    req->rq_import->imp_generation) {
-                        CDEBUG(D_HA, "%s: resend cross eviction for object: "
-                               ""LPU64":"LPU64", rc = %d.\n",
-                               req->rq_import->imp_obd->obd_name,
-                               aa->aa_oa->o_id, aa->aa_oa->o_seq, rc);
-                } else if (rc == -EINPROGRESS ||
-                    client_should_resend(aa->aa_resends, aa->aa_cli)) {
-                        rc = osc_brw_redo_request(req, aa, rc);
-                } else {
-                        CERROR("%s: too many resent retries for object: "
-                               ""LPU64":"LPU64", rc = %d.\n",
-                               req->rq_import->imp_obd->obd_name,
-                               aa->aa_oa->o_id, aa->aa_oa->o_seq, rc);
-                }
+       if (osc_recoverable_error(rc)) {
+               if (req->rq_import_generation !=
+                   req->rq_import->imp_generation) {
+                       CDEBUG(D_HA, "%s: resend cross eviction for object: "
+                              ""DOSTID", rc = %d.\n",
+                              req->rq_import->imp_obd->obd_name,
+                              POSTID(&aa->aa_oa->o_oi), rc);
+               } else if (rc == -EINPROGRESS ||
+                   client_should_resend(aa->aa_resends, aa->aa_cli)) {
+                       rc = osc_brw_redo_request(req, aa, rc);
+               } else {
+                       CERROR("%s: too many resent retries for object: "
+                              ""LPU64":"LPU64", rc = %d.\n",
+                              req->rq_import->imp_obd->obd_name,
+                              POSTID(&aa->aa_oa->o_oi), rc);
+               }
 
-                if (rc == 0)
-                        RETURN(0);
-                else if (rc == -EAGAIN || rc == -EINPROGRESS)
-                        rc = -EIO;
-        }
+               if (rc == 0)
+                       RETURN(0);
+               else if (rc == -EAGAIN || rc == -EINPROGRESS)
+                       rc = -EIO;
+       }
 
         if (aa->aa_ocapa) {
                 capa_put(aa->aa_ocapa);
@@ -2005,6 +2045,20 @@ static int brw_interpret(const struct lu_env *env,
        RETURN(rc);
 }
 
+static void brw_commit(struct ptlrpc_request *req)
+{
+       spin_lock(&req->rq_lock);
+       /* If osc_inc_unstable_pages (via osc_extent_finish) races with
+        * this called via the rq_commit_cb, I need to ensure
+        * osc_dec_unstable_pages is still called. Otherwise unstable
+        * pages may be leaked. */
+       if (req->rq_unstable)
+               osc_dec_unstable_pages(req);
+       else
+               req->rq_committed = 1;
+       spin_unlock(&req->rq_lock);
+}
+
 /**
  * Build an RPC by the list of extent @ext_list. The caller must ensure
  * that the total pages in this list are NOT over max pages per RPC.
@@ -2110,7 +2164,9 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
                GOTO(out, rc);
        }
 
+       req->rq_commit_cb = brw_commit;
        req->rq_interpret_reply = brw_interpret;
+
        if (mem_tight != 0)
                 req->rq_memalloc = 1;
 
@@ -2224,17 +2280,17 @@ static int osc_set_lock_data_with_check(struct ldlm_lock *lock,
         LASSERT(lock->l_glimpse_ast == einfo->ei_cb_gl);
 
         lock_res_and_lock(lock);
-        cfs_spin_lock(&osc_ast_guard);
+       spin_lock(&osc_ast_guard);
 
-        if (lock->l_ast_data == NULL)
-                lock->l_ast_data = data;
-        if (lock->l_ast_data == data)
-                set = 1;
+       if (lock->l_ast_data == NULL)
+               lock->l_ast_data = data;
+       if (lock->l_ast_data == data)
+               set = 1;
 
-        cfs_spin_unlock(&osc_ast_guard);
-        unlock_res_and_lock(lock);
+       spin_unlock(&osc_ast_guard);
+       unlock_res_and_lock(lock);
 
-        return set;
+       return set;
 }
 
 static int osc_set_data_with_check(struct lustre_handle *lockh,
@@ -2258,7 +2314,7 @@ static int osc_change_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm,
         struct ldlm_res_id res_id;
         struct obd_device *obd = class_exp2obd(exp);
 
-        osc_build_res_name(lsm->lsm_object_id, lsm->lsm_object_seq, &res_id);
+       ostid_build_res_name(&lsm->lsm_oi, &res_id);
         ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data);
         return 0;
 }
@@ -2274,7 +2330,7 @@ static int osc_find_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm,
         struct obd_device *obd = class_exp2obd(exp);
         int rc = 0;
 
-        osc_build_res_name(lsm->lsm_object_id, lsm->lsm_object_seq, &res_id);
+       ostid_build_res_name(&lsm->lsm_oi, &res_id);
         rc = ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data);
         if (rc == LDLM_ITER_STOP)
                 return(1);
@@ -2285,7 +2341,7 @@ static int osc_find_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm,
 
 static int osc_enqueue_fini(struct ptlrpc_request *req, struct ost_lvb *lvb,
                             obd_enqueue_update_f upcall, void *cookie,
-                            int *flags, int agl, int rc)
+                           __u64 *flags, int agl, int rc)
 {
         int intent = *flags & LDLM_FL_HAS_INTENT;
         ENTRY;
@@ -2324,7 +2380,7 @@ static int osc_enqueue_interpret(const struct lu_env *env,
         __u32 mode;
         struct ost_lvb *lvb;
         __u32 lvb_len;
-        int *flags = aa->oa_flags;
+       __u64 *flags = aa->oa_flags;
 
         /* Make a local copy of a lock handle and a mode, because aa->oa_*
          * might be freed anytime after lock upcall has been called. */
@@ -2432,7 +2488,7 @@ struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
  * is excluded from the cluster -- such scenarious make the life difficult, so
  * release locks just after they are obtained. */
 int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
-                     int *flags, ldlm_policy_data_t *policy,
+                    __u64 *flags, ldlm_policy_data_t *policy,
                      struct ost_lvb *lvb, int kms_valid,
                      obd_enqueue_update_f upcall, void *cookie,
                      struct ldlm_enqueue_info *einfo,
@@ -2542,7 +2598,7 @@ int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
         *flags &= ~LDLM_FL_BLOCK_GRANTED;
 
         rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
-                              sizeof(*lvb), lockh, async);
+                             sizeof(*lvb), LVB_T_OST, lockh, async);
         if (rqset) {
                 if (!rc) {
                         struct osc_enqueue_args *aa;
@@ -2584,9 +2640,7 @@ static int osc_enqueue(struct obd_export *exp, struct obd_info *oinfo,
         int rc;
         ENTRY;
 
-        osc_build_res_name(oinfo->oi_md->lsm_object_id,
-                           oinfo->oi_md->lsm_object_seq, &res_id);
-
+       ostid_build_res_name(&oinfo->oi_md->lsm_oi, &res_id);
         rc = osc_enqueue_base(exp, &res_id, &oinfo->oi_flags, &oinfo->oi_policy,
                               &oinfo->oi_md->lsm_oinfo[0]->loi_lvb,
                               oinfo->oi_md->lsm_oinfo[0]->loi_kms_valid,
@@ -2666,10 +2720,10 @@ static int osc_cancel_unused(struct obd_export *exp,
         struct obd_device *obd = class_exp2obd(exp);
         struct ldlm_res_id res_id, *resp = NULL;
 
-        if (lsm != NULL) {
-                resp = osc_build_res_name(lsm->lsm_object_id,
-                                          lsm->lsm_object_seq, &res_id);
-        }
+       if (lsm != NULL) {
+               ostid_build_res_name(&lsm->lsm_oi, &res_id);
+               resp = &res_id;
+       }
 
         return ldlm_cli_cancel_unused(obd->obd_namespace, resp, flags, opaque);
 }
@@ -2678,9 +2732,7 @@ static int osc_statfs_interpret(const struct lu_env *env,
                                 struct ptlrpc_request *req,
                                 struct osc_async_args *aa, int rc)
 {
-        struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
         struct obd_statfs *msfs;
-        __u64 used;
         ENTRY;
 
         if (rc == -EBADR)
@@ -2703,51 +2755,6 @@ static int osc_statfs_interpret(const struct lu_env *env,
                 GOTO(out, rc = -EPROTO);
         }
 
-        /* Reinitialize the RDONLY and DEGRADED flags at the client
-         * on each statfs, so they don't stay set permanently. */
-        cfs_spin_lock(&cli->cl_oscc.oscc_lock);
-
-        if (unlikely(msfs->os_state & OS_STATE_DEGRADED))
-                cli->cl_oscc.oscc_flags |= OSCC_FLAG_DEGRADED;
-        else if (unlikely(cli->cl_oscc.oscc_flags & OSCC_FLAG_DEGRADED))
-                cli->cl_oscc.oscc_flags &= ~OSCC_FLAG_DEGRADED;
-
-        if (unlikely(msfs->os_state & OS_STATE_READONLY))
-                cli->cl_oscc.oscc_flags |= OSCC_FLAG_RDONLY;
-        else if (unlikely(cli->cl_oscc.oscc_flags & OSCC_FLAG_RDONLY))
-                cli->cl_oscc.oscc_flags &= ~OSCC_FLAG_RDONLY;
-
-        /* Add a bit of hysteresis so this flag isn't continually flapping,
-         * and ensure that new files don't get extremely fragmented due to
-         * only a small amount of available space in the filesystem.
-         * We want to set the NOSPC flag when there is less than ~0.1% free
-         * and clear it when there is at least ~0.2% free space, so:
-         *                   avail < ~0.1% max          max = avail + used
-         *            1025 * avail < avail + used       used = blocks - free
-         *            1024 * avail < used
-         *            1024 * avail < blocks - free
-         *                   avail < ((blocks - free) >> 10)
-         *
-         * On very large disk, say 16TB 0.1% will be 16 GB. We don't want to
-         * lose that amount of space so in those cases we report no space left
-         * if their is less than 1 GB left.                             */
-        used = min_t(__u64,(msfs->os_blocks - msfs->os_bfree) >> 10, 1 << 30);
-        if (unlikely(((cli->cl_oscc.oscc_flags & OSCC_FLAG_NOSPC) == 0) &&
-                     ((msfs->os_ffree < 32) || (msfs->os_bavail < used))))
-                cli->cl_oscc.oscc_flags |= OSCC_FLAG_NOSPC;
-        else if (unlikely(((cli->cl_oscc.oscc_flags & OSCC_FLAG_NOSPC) != 0) &&
-                          (msfs->os_ffree > 64) &&
-                          (msfs->os_bavail > (used << 1)))) {
-                cli->cl_oscc.oscc_flags &= ~(OSCC_FLAG_NOSPC |
-                                             OSCC_FLAG_NOSPC_BLK);
-        }
-
-        if (unlikely(((cli->cl_oscc.oscc_flags & OSCC_FLAG_NOSPC) != 0) &&
-                     (msfs->os_bavail < used)))
-                cli->cl_oscc.oscc_flags |= OSCC_FLAG_NOSPC_BLK;
-
-        cfs_spin_unlock(&cli->cl_oscc.oscc_lock);
-
         *aa->aa_oi->oi_osfs = *msfs;
 out:
         rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
@@ -2810,10 +2817,10 @@ static int osc_statfs(const struct lu_env *env, struct obd_export *exp,
 
         /*Since the request might also come from lprocfs, so we need
          *sync this with client_disconnect_export Bug15684*/
-        cfs_down_read(&obd->u.cli.cl_sem);
+       down_read(&obd->u.cli.cl_sem);
         if (obd->u.cli.cl_import)
                 imp = class_import_get(obd->u.cli.cl_import);
-        cfs_up_read(&obd->u.cli.cl_sem);
+       up_read(&obd->u.cli.cl_sem);
         if (!imp)
                 RETURN(-ENODEV);
 
@@ -2902,18 +2909,18 @@ static int osc_getstripe(struct lov_stripe_md *lsm, struct lov_user_md *lump)
                 if (!lumk)
                         RETURN(-ENOMEM);
 
-                if (lum.lmm_magic == LOV_USER_MAGIC_V1)
-                        lmm_objects = &(((struct lov_user_md_v1 *)lumk)->lmm_objects[0]);
-                else
-                        lmm_objects = &(lumk->lmm_objects[0]);
-                lmm_objects->l_object_id = lsm->lsm_object_id;
-        } else {
-                lum_size = lov_mds_md_size(0, lum.lmm_magic);
-                lumk = &lum;
-        }
+               if (lum.lmm_magic == LOV_USER_MAGIC_V1)
+                       lmm_objects =
+                           &(((struct lov_user_md_v1 *)lumk)->lmm_objects[0]);
+               else
+                       lmm_objects = &(lumk->lmm_objects[0]);
+               lmm_objects->l_ost_oi = lsm->lsm_oi;
+       } else {
+               lum_size = lov_mds_md_size(0, lum.lmm_magic);
+               lumk = &lum;
+       }
 
-        lumk->lmm_object_id = lsm->lsm_object_id;
-        lumk->lmm_object_seq = lsm->lsm_object_seq;
+       lumk->lmm_oi = lsm->lsm_oi;
         lumk->lmm_stripe_count = 1;
 
         if (cfs_copy_to_user(lump, lumk, lum_size))
@@ -3109,40 +3116,6 @@ static int osc_get_info(const struct lu_env *env, struct obd_export *exp,
         RETURN(-EINVAL);
 }
 
-static int osc_setinfo_mds_connect_import(struct obd_import *imp)
-{
-        struct llog_ctxt *ctxt;
-        int rc = 0;
-        ENTRY;
-
-        ctxt = llog_get_context(imp->imp_obd, LLOG_MDS_OST_ORIG_CTXT);
-        if (ctxt) {
-                rc = llog_initiator_connect(ctxt);
-                llog_ctxt_put(ctxt);
-        } else {
-                /* XXX return an error? skip setting below flags? */
-        }
-
-        cfs_spin_lock(&imp->imp_lock);
-        imp->imp_server_timeout = 1;
-        imp->imp_pingable = 1;
-        cfs_spin_unlock(&imp->imp_lock);
-        CDEBUG(D_RPCTRACE, "pinging OST %s\n", obd2cli_tgt(imp->imp_obd));
-
-        RETURN(rc);
-}
-
-static int osc_setinfo_mds_conn_interpret(const struct lu_env *env,
-                                          struct ptlrpc_request *req,
-                                          void *aa, int rc)
-{
-        ENTRY;
-        if (rc != 0)
-                RETURN(rc);
-
-        RETURN(osc_setinfo_mds_connect_import(req->rq_import));
-}
-
 static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
                               obd_count keylen, void *key, obd_count vallen,
                               void *val, struct ptlrpc_request_set *set)
@@ -3156,32 +3129,6 @@ static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
 
         OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_SHUTDOWN, 10);
 
-        if (KEY_IS(KEY_NEXT_ID)) {
-                obd_id new_val;
-                struct osc_creator *oscc = &obd->u.cli.cl_oscc;
-
-                if (vallen != sizeof(obd_id))
-                        RETURN(-ERANGE);
-                if (val == NULL)
-                        RETURN(-EINVAL);
-
-                if (vallen != sizeof(obd_id))
-                        RETURN(-EINVAL);
-
-                /* avoid race between allocate new object and set next id
-                 * from ll_sync thread */
-                cfs_spin_lock(&oscc->oscc_lock);
-                new_val = *((obd_id*)val) + 1;
-                if (new_val > oscc->oscc_next_id)
-                        oscc->oscc_next_id = new_val;
-                cfs_spin_unlock(&oscc->oscc_lock);
-                CDEBUG(D_HA, "%s: set oscc_next_id = "LPU64"\n",
-                       exp->exp_obd->obd_name,
-                       obd->u.cli.cl_oscc.oscc_next_id);
-
-                RETURN(0);
-        }
-
         if (KEY_IS(KEY_CHECKSUM)) {
                 if (vallen != sizeof(int))
                         RETURN(-EINVAL);
@@ -3199,6 +3146,33 @@ static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
                 RETURN(0);
         }
 
+       if (KEY_IS(KEY_CACHE_SET)) {
+               struct client_obd *cli = &obd->u.cli;
+
+               LASSERT(cli->cl_cache == NULL); /* only once */
+               cli->cl_cache = (struct cl_client_cache *)val;
+               cfs_atomic_inc(&cli->cl_cache->ccc_users);
+               cli->cl_lru_left = &cli->cl_cache->ccc_lru_left;
+
+               /* add this osc into entity list */
+               LASSERT(cfs_list_empty(&cli->cl_lru_osc));
+               spin_lock(&cli->cl_cache->ccc_lru_lock);
+               cfs_list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru);
+               spin_unlock(&cli->cl_cache->ccc_lru_lock);
+
+               RETURN(0);
+       }
+
+       if (KEY_IS(KEY_CACHE_LRU_SHRINK)) {
+               struct client_obd *cli = &obd->u.cli;
+               int nr = cfs_atomic_read(&cli->cl_lru_in_list) >> 1;
+               int target = *(int *)val;
+
+               nr = osc_lru_shrink(cli, min(nr, target));
+               *(int *)val -= nr;
+               RETURN(0);
+       }
+
         if (!set && !KEY_IS(KEY_GRANT_SHRINK))
                 RETURN(-EINVAL);
 
@@ -3209,38 +3183,31 @@ static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
            Even if something bad goes through, we'd get a -EINVAL from OST
            anyway. */
 
-        if (KEY_IS(KEY_GRANT_SHRINK))
-                req = ptlrpc_request_alloc(imp, &RQF_OST_SET_GRANT_INFO);
-        else
-                req = ptlrpc_request_alloc(imp, &RQF_OBD_SET_INFO);
-
-        if (req == NULL)
-                RETURN(-ENOMEM);
-
-        req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
-                             RCL_CLIENT, keylen);
-        req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
-                             RCL_CLIENT, vallen);
-        rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
-        if (rc) {
-                ptlrpc_request_free(req);
-                RETURN(rc);
-        }
+       req = ptlrpc_request_alloc(imp, KEY_IS(KEY_GRANT_SHRINK) ?
+                                               &RQF_OST_SET_GRANT_INFO :
+                                               &RQF_OBD_SET_INFO);
+       if (req == NULL)
+               RETURN(-ENOMEM);
+
+       req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
+                            RCL_CLIENT, keylen);
+       if (!KEY_IS(KEY_GRANT_SHRINK))
+               req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
+                                    RCL_CLIENT, vallen);
+       rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
 
-        tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
-        memcpy(tmp, key, keylen);
-        tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
+       tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
+       memcpy(tmp, key, keylen);
+       tmp = req_capsule_client_get(&req->rq_pill, KEY_IS(KEY_GRANT_SHRINK) ?
+                                                       &RMF_OST_BODY :
+                                                       &RMF_SETINFO_VAL);
         memcpy(tmp, val, vallen);
 
-        if (KEY_IS(KEY_MDS_CONN)) {
-                struct osc_creator *oscc = &obd->u.cli.cl_oscc;
-
-                oscc->oscc_oa.o_seq = (*(__u32 *)val);
-                oscc->oscc_oa.o_valid |= OBD_MD_FLGROUP;
-                LASSERT_SEQ_IS_MDT(oscc->oscc_oa.o_seq);
-                req->rq_no_delay = req->rq_no_resend = 1;
-                req->rq_interpret_reply = osc_setinfo_mds_conn_interpret;
-        } else if (KEY_IS(KEY_GRANT_SHRINK)) {
+       if (KEY_IS(KEY_GRANT_SHRINK)) {
                 struct osc_grant_args *aa;
                 struct obdo *oa;
 
@@ -3268,101 +3235,31 @@ static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
 }
 
 
-static struct llog_operations osc_size_repl_logops = {
-        lop_cancel: llog_obd_repl_cancel
-};
-
-static struct llog_operations osc_mds_ost_orig_logops;
-
-static int __osc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
-                           struct obd_device *tgt, struct llog_catid *catid)
-{
-        int rc;
-        ENTRY;
-
-        rc = llog_setup(obd, &obd->obd_olg, LLOG_MDS_OST_ORIG_CTXT, tgt, 1,
-                        &catid->lci_logid, &osc_mds_ost_orig_logops);
-        if (rc) {
-                CERROR("failed LLOG_MDS_OST_ORIG_CTXT\n");
-                GOTO(out, rc);
-        }
-
-        rc = llog_setup(obd, &obd->obd_olg, LLOG_SIZE_REPL_CTXT, tgt, 1,
-                        NULL, &osc_size_repl_logops);
-        if (rc) {
-                struct llog_ctxt *ctxt =
-                        llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
-                if (ctxt)
-                        llog_cleanup(ctxt);
-                CERROR("failed LLOG_SIZE_REPL_CTXT\n");
-        }
-        GOTO(out, rc);
-out:
-        if (rc) {
-                CERROR("osc '%s' tgt '%s' catid %p rc=%d\n",
-                       obd->obd_name, tgt->obd_name, catid, rc);
-                CERROR("logid "LPX64":0x%x\n",
-                       catid->lci_logid.lgl_oid, catid->lci_logid.lgl_ogen);
-        }
-        return rc;
-}
-
 static int osc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
                          struct obd_device *disk_obd, int *index)
 {
-        struct llog_catid catid;
-        static char name[32] = CATLIST;
-        int rc;
-        ENTRY;
-
-        LASSERT(olg == &obd->obd_olg);
-
-        cfs_mutex_lock(&olg->olg_cat_processing);
-        rc = llog_get_cat_list(disk_obd, name, *index, 1, &catid);
-        if (rc) {
-                CERROR("rc: %d\n", rc);
-                GOTO(out, rc);
-        }
-
-        CDEBUG(D_INFO, "%s: Init llog for %d - catid "LPX64"/"LPX64":%x\n",
-               obd->obd_name, *index, catid.lci_logid.lgl_oid,
-               catid.lci_logid.lgl_oseq, catid.lci_logid.lgl_ogen);
-
-        rc = __osc_llog_init(obd, olg, disk_obd, &catid);
-        if (rc) {
-                CERROR("rc: %d\n", rc);
-                GOTO(out, rc);
-        }
-
-        rc = llog_put_cat_list(disk_obd, name, *index, 1, &catid);
-        if (rc) {
-                CERROR("rc: %d\n", rc);
-                GOTO(out, rc);
-        }
-
- out:
-        cfs_mutex_unlock(&olg->olg_cat_processing);
-
-        return rc;
+       /* this code is not supposed to be used with LOD/OSP
+        * to be removed soon */
+       LBUG();
+       return 0;
 }
 
 static int osc_llog_finish(struct obd_device *obd, int count)
 {
-        struct llog_ctxt *ctxt;
-        int rc = 0, rc2 = 0;
-        ENTRY;
+       struct llog_ctxt *ctxt;
 
-        ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
-        if (ctxt)
-                rc = llog_cleanup(ctxt);
+       ENTRY;
 
-        ctxt = llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
-        if (ctxt)
-                rc2 = llog_cleanup(ctxt);
-        if (!rc)
-                rc = rc2;
+       ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
+       if (ctxt) {
+               llog_cat_close(NULL, ctxt->loc_handle);
+               llog_cleanup(NULL, ctxt);
+       }
 
-        RETURN(rc);
+       ctxt = llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
+       if (ctxt)
+               llog_cleanup(NULL, ctxt);
+       RETURN(0);
 }
 
 static int osc_reconnect(const struct lu_env *env,
@@ -3378,7 +3275,7 @@ static int osc_reconnect(const struct lu_env *env,
 
                 client_obd_list_lock(&cli->cl_loi_list_lock);
                 data->ocd_grant = (cli->cl_avail_grant + cli->cl_dirty) ?:
-                                2 * cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT;
+                               2 * cli_brw_size(obd);
                 lost_grant = cli->cl_lost_grant;
                 cli->cl_lost_grant = 0;
                 client_obd_list_unlock(&cli->cl_loi_list_lock);
@@ -3445,14 +3342,6 @@ static int osc_import_event(struct obd_device *obd,
 
         switch (event) {
         case IMP_EVENT_DISCON: {
-                /* Only do this on the MDS OSC's */
-                if (imp->imp_server_timeout) {
-                        struct osc_creator *oscc = &obd->u.cli.cl_oscc;
-
-                        cfs_spin_lock(&oscc->oscc_lock);
-                        oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
-                        cfs_spin_unlock(&oscc->oscc_lock);
-                }
                 cli = &obd->u.cli;
                 client_obd_list_lock(&cli->cl_loi_list_lock);
                 cli->cl_avail_grant = 0;
@@ -3484,15 +3373,6 @@ static int osc_import_event(struct obd_device *obd,
                 break;
         }
         case IMP_EVENT_ACTIVE: {
-                /* Only do this on the MDS OSC's */
-                if (imp->imp_server_timeout) {
-                        struct osc_creator *oscc = &obd->u.cli.cl_oscc;
-
-                        cfs_spin_lock(&oscc->oscc_lock);
-                        oscc->oscc_flags &= ~(OSCC_FLAG_NOSPC |
-                                              OSCC_FLAG_NOSPC_BLK);
-                        cfs_spin_unlock(&oscc->oscc_lock);
-                }
                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
                 break;
         }
@@ -3578,7 +3458,7 @@ int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 
        handler = ptlrpcd_alloc_work(cli->cl_import, brw_queue_work, cli);
        if (IS_ERR(handler))
-               GOTO(out_client_setup, PTR_ERR(handler));
+               GOTO(out_client_setup, rc = PTR_ERR(handler));
        cli->cl_writeback_work = handler;
 
        rc = osc_quota_setup(obd);
@@ -3593,7 +3473,6 @@ int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
                ptlrpc_lprocfs_register_obd(obd);
        }
 
-       oscc_init(obd);
        /* We need to allocate a few requests more, because
         * brw_interpret tries to create new requests before freeing
         * previous ones, Ideally we want to have 2x max_rpcs_in_flight
@@ -3629,9 +3508,9 @@ static int osc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
                 CDEBUG(D_HA, "Deactivating import %s\n", obd->obd_name);
                 /* ptlrpc_abort_inflight to stop an mds_lov_synchronize */
                 ptlrpc_deactivate_import(imp);
-                cfs_spin_lock(&imp->imp_lock);
-                imp->imp_pingable = 0;
-                cfs_spin_unlock(&imp->imp_lock);
+               spin_lock(&imp->imp_lock);
+               imp->imp_pingable = 0;
+               spin_unlock(&imp->imp_lock);
                 break;
         }
         case OBD_CLEANUP_EXPORTS: {
@@ -3664,9 +3543,21 @@ static int osc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 
 int osc_cleanup(struct obd_device *obd)
 {
-        int rc;
+       struct client_obd *cli = &obd->u.cli;
+       int rc;
 
-        ENTRY;
+       ENTRY;
+
+       /* lru cleanup */
+       if (cli->cl_cache != NULL) {
+               LASSERT(cfs_atomic_read(&cli->cl_cache->ccc_users) > 0);
+               spin_lock(&cli->cl_cache->ccc_lru_lock);
+               cfs_list_del_init(&cli->cl_lru_osc);
+               spin_unlock(&cli->cl_cache->ccc_lru_lock);
+               cli->cl_lru_left = NULL;
+               cfs_atomic_dec(&cli->cl_cache->ccc_users);
+               cli->cl_cache = NULL;
+       }
 
         /* free memory of osc quota cache */
         osc_quota_cleanup(obd);
@@ -3715,9 +3606,7 @@ struct obd_ops osc_obd_ops = {
         .o_statfs_async         = osc_statfs_async,
         .o_packmd               = osc_packmd,
         .o_unpackmd             = osc_unpackmd,
-        .o_precreate            = osc_precreate,
         .o_create               = osc_create,
-        .o_create_async         = osc_create_async,
         .o_destroy              = osc_destroy,
         .o_getattr              = osc_getattr,
         .o_getattr_async        = osc_getattr_async,
@@ -3740,12 +3629,11 @@ struct obd_ops osc_obd_ops = {
         .o_process_config       = osc_process_config,
         .o_quotactl             = osc_quotactl,
         .o_quotacheck           = osc_quotacheck,
-        .o_quota_adjust_qunit   = osc_quota_adjust_qunit,
 };
 
 extern struct lu_kmem_descr osc_caches[];
-extern cfs_spinlock_t       osc_ast_guard;
-extern cfs_lock_class_key_t osc_ast_guard_class;
+extern spinlock_t osc_ast_guard;
+extern struct lock_class_key osc_ast_guard_class;
 
 int __init osc_init(void)
 {
@@ -3769,16 +3657,10 @@ int __init osc_init(void)
                 RETURN(rc);
         }
 
-        cfs_spin_lock_init(&osc_ast_guard);
-        cfs_lockdep_set_class(&osc_ast_guard, &osc_ast_guard_class);
+       spin_lock_init(&osc_ast_guard);
+       lockdep_set_class(&osc_ast_guard, &osc_ast_guard_class);
 
-        osc_mds_ost_orig_logops = llog_lvfs_ops;
-        osc_mds_ost_orig_logops.lop_setup = llog_obd_origin_setup;
-        osc_mds_ost_orig_logops.lop_cleanup = llog_obd_origin_cleanup;
-        osc_mds_ost_orig_logops.lop_add = llog_obd_origin_add;
-        osc_mds_ost_orig_logops.lop_connect = llog_origin_connect;
-
-        RETURN(rc);
+       RETURN(rc);
 }
 
 #ifdef __KERNEL__