Whamcloud - gitweb
b=19387 integrate LST into acc-sm
[fs/lustre-release.git] / lustre / ost / ost_handler.c
index a2fd9b9..9c20008 100644 (file)
@@ -106,70 +106,140 @@ static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
                        struct obd_trans_info *oti)
 {
         struct ost_body *body, *repbody;
-        __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
         struct lustre_capa *capa = NULL;
         int rc;
         ENTRY;
 
-        body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
-                                  lustre_swab_ost_body);
+        /* Get the request body */
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
         if (body == NULL)
                 RETURN(-EFAULT);
 
         if (body->oa.o_id == 0)
                 RETURN(-EPROTO);
 
-        if (lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1)) {
+        /* If there's a DLM request, cancel the locks mentioned in it*/
+        if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
                 struct ldlm_request *dlm;
-                dlm = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*dlm),
-                                         lustre_swab_ldlm_request);
+
+                dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
                 if (dlm == NULL)
                         RETURN (-EFAULT);
                 ldlm_request_cancel(req, dlm, 0);
         }
 
-        if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
-                capa = lustre_unpack_incoming_capa(req, REQ_REC_OFF + 2);
+        /* If there's a capability, get it */
+        if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+                capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
+                if (capa == NULL) {
+                        CERROR("Missing capability for OST DESTROY");
+                        RETURN (-EFAULT);
+                }
+        }
 
-        rc = lustre_pack_reply(req, 2, size, NULL);
+        /* Prepare the reply */
+        rc = req_capsule_server_pack(&req->rq_pill);
         if (rc)
                 RETURN(rc);
 
+        /* Get the log cancellation cookie */
         if (body->oa.o_valid & OBD_MD_FLCOOKIE)
                 oti->oti_logcookies = &body->oa.o_lcookie;
-        repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
-                                 sizeof(*repbody));
+
+        /* Finish the reply */
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
+
+        /* Do the destroy and set the reply status accordingly  */
         req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL, capa);
         RETURN(0);
 }
 
+/**
+ * Helper function for getting server side [start, start+count] DLM lock
+ * if asked by client.
+ */
+static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
+                        __u64 start, __u64 count, struct lustre_handle *lh,
+                        int mode, int flags)
+{
+        struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0} };
+        ldlm_policy_data_t policy;
+        __u64 end = start + count;
+
+        ENTRY;
+
+        LASSERT(!lustre_handle_is_used(lh));
+        LASSERT((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) ==
+                (OBD_MD_FLID | OBD_MD_FLGROUP));
+
+        if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
+            !(oa->o_flags & OBD_FL_SRVLOCK))
+                RETURN(0);
+
+        CDEBUG(D_INODE, "OST-side extent lock.\n");
+
+        policy.l_extent.start = start & CFS_PAGE_MASK;
+
+        /* If ->o_blocks is EOF it means "lock till the end of the
+         * file". Otherwise, it's size of a hole being punched (in bytes) */
+        if (count == OBD_OBJECT_EOF || end < start)
+                policy.l_extent.end = OBD_OBJECT_EOF;
+        else
+                policy.l_extent.end = end | ~CFS_PAGE_MASK;
+
+        RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
+                                      LDLM_EXTENT, &policy, mode, &flags,
+                                      ldlm_blocking_ast, ldlm_completion_ast,
+                                      ldlm_glimpse_ast, NULL, 0, NULL, lh));
+}
+
+/* Helper function: release lock, if any. */
+static void ost_lock_put(struct obd_export *exp,
+                         struct lustre_handle *lh, int mode)
+{
+        ENTRY;
+        if (lustre_handle_is_used(lh))
+                ldlm_lock_decref(lh, mode);
+        EXIT;
+}
+
 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 {
         struct ost_body *body, *repbody;
         struct obd_info oinfo = { { { 0 } } };
-        __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
+        struct lustre_handle lh = { 0 };
         int rc;
         ENTRY;
 
-        body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
-                                  lustre_swab_ost_body);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
         if (body == NULL)
                 RETURN(-EFAULT);
 
-        rc = lustre_pack_reply(req, 2, size, NULL);
+        rc = req_capsule_server_pack(&req->rq_pill);
         if (rc)
                 RETURN(rc);
 
-        repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
-                                 sizeof(*repbody));
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
         repbody->oa = body->oa;
 
+        rc = ost_lock_get(exp, &body->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
+        if (rc)
+                RETURN(rc);
+
         oinfo.oi_oa = &repbody->oa;
-        if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
-                oinfo.oi_capa = lustre_unpack_incoming_capa(req,
-                                                   REQ_REC_OFF + 1);
+        if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
+                oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
+                                                       &RMF_CAPA1);
+                if (oinfo.oi_capa == NULL) {
+                        CERROR("Missing capability for OST GETATTR");
+                        RETURN (-EFAULT);
+                }
+        }
+
         req->rq_status = obd_getattr(exp, &oinfo);
+        ost_lock_put(exp, &lh, LCK_PR);
+
         ost_drop_id(exp, &repbody->oa);
         RETURN(0);
 }
@@ -177,20 +247,17 @@ static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 static int ost_statfs(struct ptlrpc_request *req)
 {
         struct obd_statfs *osfs;
-        __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) };
         int rc;
         ENTRY;
 
-        rc = lustre_pack_reply(req, 2, size, NULL);
+        rc = req_capsule_server_pack(&req->rq_pill);
         if (rc)
                 RETURN(rc);
 
-        osfs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*osfs));
+        osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
 
         req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
-                                    cfs_time_current_64() - HZ, 0);
-        if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
-                osfs->os_bfree = osfs->os_bavail = 64;
+                                    cfs_time_current_64() - CFS_HZ, 0);
         if (req->rq_status != 0)
                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 
@@ -201,21 +268,18 @@ static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
                       struct obd_trans_info *oti)
 {
         struct ost_body *body, *repbody;
-        __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
         int rc;
         ENTRY;
 
-        body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
-                                  lustre_swab_ost_body);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
         if (body == NULL)
                 RETURN(-EFAULT);
 
-        rc = lustre_pack_reply(req, 2, size, NULL);
+        rc = req_capsule_server_pack(&req->rq_pill);
         if (rc)
                 RETURN(rc);
 
-        repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
-                                 sizeof(*repbody));
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
         oti->oti_logcookies = &repbody->oa.o_lcookie;
 
@@ -224,86 +288,21 @@ static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
         RETURN(0);
 }
 
-/*
- * Helper function for ost_punch(): if asked by client, acquire [size, EOF]
- * lock on the file being truncated.
- */
-static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa,
-                              struct lustre_handle *lh)
-{
-        int flags;
-        struct ldlm_res_id res_id;
-        ldlm_policy_data_t policy;
-        __u64 start;
-        __u64 finis;
-
-        ENTRY;
-
-        osc_build_res_name(oa->o_id, oa->o_gr, &res_id);
-        LASSERT(!lustre_handle_is_used(lh));
-
-        if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
-            !(oa->o_flags & OBD_FL_TRUNCLOCK))
-                RETURN(0);
-
-        CDEBUG(D_INODE, "OST-side truncate lock.\n");
-
-        start = oa->o_size;
-        finis = start + oa->o_blocks;
-
-        /*
-         * standard truncate optimization: if file body is completely
-         * destroyed, don't send data back to the server.
-         */
-        flags = (start == 0) ? LDLM_AST_DISCARD_DATA : 0;
-
-        policy.l_extent.start = start & CFS_PAGE_MASK;
-
-        /*
-         * If ->o_blocks is EOF it means "lock till the end of the
-         * file". Otherwise, it's size of a hole being punched (in bytes)
-         */
-        if (oa->o_blocks == OBD_OBJECT_EOF || finis < start)
-                policy.l_extent.end = OBD_OBJECT_EOF;
-        else
-                policy.l_extent.end = finis | ~CFS_PAGE_MASK;
-
-        RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
-                                      LDLM_EXTENT, &policy, LCK_PW, &flags,
-                                      ldlm_blocking_ast, ldlm_completion_ast,
-                                      ldlm_glimpse_ast, NULL, 0, NULL,
-                                      NULL, lh));
-}
-
-/*
- * Helper function for ost_punch(): release lock acquired by
- * ost_punch_lock_get(), if any.
- */
-static void ost_punch_lock_put(struct obd_export *exp, struct obdo *oa,
-                               struct lustre_handle *lh)
-{
-        ENTRY;
-        if (lustre_handle_is_used(lh))
-                ldlm_lock_decref(lh, LCK_PW);
-        EXIT;
-}
-
 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
                      struct obd_trans_info *oti)
 {
         struct obd_info oinfo = { { { 0 } } };
         struct ost_body *body, *repbody;
-        __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
-        int rc;
+        int rc, flags = 0;
         struct lustre_handle lh = {0,};
         ENTRY;
 
         /* check that we do support OBD_CONNECT_TRUNCLOCK. */
         CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
 
-        /* ost_body is varified and swabbed in ost_hpreq_handler() */
-        body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
-        LASSERT(body != NULL);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+        if (body == NULL)
+                RETURN(-EFAULT);
 
         oinfo.oi_oa = &body->oa;
         oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
@@ -313,28 +312,38 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
                 RETURN(-EPROTO);
 
-        rc = lustre_pack_reply(req, 2, size, NULL);
+        rc = req_capsule_server_pack(&req->rq_pill);
         if (rc)
                 RETURN(rc);
 
-        repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
-                                 sizeof(*repbody));
-        rc = ost_punch_lock_get(exp, oinfo.oi_oa, &lh);
+        /* standard truncate optimization: if file body is completely
+         * destroyed, don't send data back to the server. */
+        if (oinfo.oi_oa->o_size == 0)
+                flags |= LDLM_AST_DISCARD_DATA;
+
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+        rc = ost_lock_get(exp, oinfo.oi_oa, oinfo.oi_oa->o_size,
+                          oinfo.oi_oa->o_blocks, &lh, LCK_PW, flags);
         if (rc == 0) {
                 if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
-                    oinfo.oi_oa->o_flags == OBD_FL_TRUNCLOCK)
+                    oinfo.oi_oa->o_flags == OBD_FL_SRVLOCK)
                         /*
-                         * If OBD_FL_TRUNCLOCK is the only bit set in
+                         * If OBD_FL_SRVLOCK is the only bit set in
                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
                          * through filter_setattr() to filter_iocontrol().
                          */
                         oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
 
-                if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
-                        oinfo.oi_capa = lustre_unpack_incoming_capa(req,
-                                                           REQ_REC_OFF + 1);
+                if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
+                        oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
+                                                               &RMF_CAPA1);
+                        if (oinfo.oi_capa == NULL) {
+                                CERROR("Missing capability for OST PUNCH");
+                                RETURN (-EFAULT);
+                        }
+                }
                 req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
-                ost_punch_lock_put(exp, oinfo.oi_oa, &lh);
+                ost_lock_put(exp, &lh, LCK_PW);
         }
         repbody->oa = *oinfo.oi_oa;
         ost_drop_id(exp, &repbody->oa);
@@ -345,24 +354,26 @@ static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
 {
         struct ost_body *body, *repbody;
         struct lustre_capa *capa = NULL;
-        __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
         int rc;
         ENTRY;
 
-        body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
-                                  lustre_swab_ost_body);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
         if (body == NULL)
                 RETURN(-EFAULT);
 
-        if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
-                capa = lustre_unpack_incoming_capa(req, REQ_REC_OFF + 1);
+        if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+                capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
+                if (capa == NULL) {
+                        CERROR("Missing capability for OST SYNC");
+                        RETURN (-EFAULT);
+                }
+        }
 
-        rc = lustre_pack_reply(req, 2, size, NULL);
+        rc = req_capsule_server_pack(&req->rq_pill);
         if (rc)
                 RETURN(rc);
 
-        repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
-                                 sizeof(*repbody));
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
         req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
                                   repbody->oa.o_blocks, capa);
@@ -374,28 +385,30 @@ static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
                        struct obd_trans_info *oti)
 {
         struct ost_body *body, *repbody;
-        __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
         int rc;
         struct obd_info oinfo = { { { 0 } } };
         ENTRY;
 
-        body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
-                                  lustre_swab_ost_body);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
         if (body == NULL)
                 RETURN(-EFAULT);
 
-        rc = lustre_pack_reply(req, 2, size, NULL);
+        rc = req_capsule_server_pack(&req->rq_pill);
         if (rc)
                 RETURN(rc);
 
-        repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
-                                 sizeof(*repbody));
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
         repbody->oa = body->oa;
 
         oinfo.oi_oa = &repbody->oa;
-        if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
-                oinfo.oi_capa = lustre_unpack_incoming_capa(req,
-                                                   REQ_REC_OFF + 1);
+        if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
+                oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
+                                                       &RMF_CAPA1);
+                if (oinfo.oi_capa == NULL) {
+                        CERROR("Missing capability for OST SETATTR");
+                        RETURN (-EFAULT);
+                }
+        }
         req->rq_status = obd_setattr(exp, &oinfo, oti);
         ost_drop_id(exp, &repbody->oa);
         RETURN(0);
@@ -461,10 +474,10 @@ static int ost_brw_lock_get(int mode, struct obd_export *exp,
         if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
                 RETURN(0);
 
-        /* EXPENSIVE ASSERTION */
         for (i = 1; i < nrbufs; i ++)
-                LASSERT((nb[0].flags & OBD_BRW_SRVLOCK) ==
-                        (nb[i].flags & OBD_BRW_SRVLOCK));
+                if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
+                    (nb[i].flags & OBD_BRW_SRVLOCK))
+                        RETURN(-EFAULT);
 
         policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
         policy.l_extent.end   = (nb[nrbufs - 1].offset +
@@ -473,8 +486,7 @@ static int ost_brw_lock_get(int mode, struct obd_export *exp,
         RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
                                       LDLM_EXTENT, &policy, mode, &flags,
                                       ldlm_blocking_ast, ldlm_completion_ast,
-                                      ldlm_glimpse_ast, NULL, 0, NULL,
-                                      NULL, lh));
+                                      ldlm_glimpse_ast, NULL, 0, NULL, lh));
 }
 
 static void ost_brw_lock_put(int mode,
@@ -619,7 +631,6 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
         struct lustre_capa *capa = NULL;
         struct l_wait_info lwi;
         struct lustre_handle lockh = { 0 };
-        __u32  size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
         int niocount, npages, nob = 0, rc, i;
         int no_reply = 0;
         ENTRY;
@@ -633,10 +644,10 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 
         /* Check if there is eviction in progress, and if so, wait for it to
          * finish */
-        if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
+        if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
-                        !atomic_read(&exp->exp_obd->obd_evict_inprogress),
+                        !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
                         &lwi);
         }
         if (exp->exp_failed)
@@ -644,21 +655,33 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
 
         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
          * ost_rw_hpreq_check(). */
-        body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
-        LASSERT(body != NULL);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+        if (body == NULL)
+                GOTO(out, rc = -EFAULT);
 
-        ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, sizeof(*ioo));
-        LASSERT(ioo != NULL);
+        /*
+         * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
+         * would be useful here and wherever we get &RMF_OBD_IOOBJ and
+         * &RMF_NIOBUF_REMOTE.
+         */
+        ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
+        if (ioo == NULL)
+                GOTO(out, rc = -EFAULT);
 
         niocount = ioo->ioo_bufcnt;
-        remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
-                                   niocount * sizeof(*remote_nb));
-        LASSERT(remote_nb != NULL);
+        remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
+        if (remote_nb == NULL)
+                GOTO(out, rc = -EFAULT);
 
-        if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
-                capa = lustre_unpack_incoming_capa(req, REQ_REC_OFF + 3);
+        if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+                capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
+                if (capa == NULL) {
+                        CERROR("Missing capability for OST BRW READ");
+                        GOTO(out, rc = -EFAULT);
+                }
+        }
 
-        rc = lustre_pack_reply(req, 2, size, NULL);
+        rc = req_capsule_server_pack(&req->rq_pill);
         if (rc)
                 GOTO(out, rc);
 
@@ -698,7 +721,11 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
         if (desc == NULL)
                 GOTO(out_lock, rc = -ENOMEM);
 
-        ost_rw_prolong_locks(req, ioo, remote_nb, &body->oa, LCK_PW | LCK_PR);
+        if (!lustre_handle_is_used(&lockh))
+                /* no needs to try to prolong lock if server is asked
+                 * to handle locking (= OBD_BRW_SRVLOCK) */
+                ost_rw_prolong_locks(req, ioo, remote_nb, &body->oa,
+                                     LCK_PW | LCK_PR);
 
         nob = 0;
         for (i = 0; i < npages; i++) {
@@ -744,13 +771,13 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
         if (rc == 0) {
                 /* Check if there is eviction in progress, and if so, wait for
                  * it to finish */
-                if (unlikely(atomic_read(&exp->exp_obd->
-                                                obd_evict_inprogress))) {
+                if (unlikely(cfs_atomic_read(&exp->exp_obd->
+                                             obd_evict_inprogress))) {
                         lwi = LWI_INTR(NULL, NULL);
                         rc = l_wait_event(exp->exp_obd->
-                                                obd_evict_inprogress_waitq,
-                                          !atomic_read(&exp->exp_obd->
-                                                        obd_evict_inprogress),
+                                          obd_evict_inprogress_waitq,
+                                          !cfs_atomic_read(&exp->exp_obd->
+                                          obd_evict_inprogress),
                                           &lwi);
                 }
                 /* Check if client was evicted or tried to reconnect already */
@@ -767,7 +794,7 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
                         do {
                                 long timeoutl = req->rq_deadline -
                                         cfs_time_current_sec();
-                                cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
+                                cfs_duration_t timeout = timeoutl <= 0 ?
                                         CFS_TICK : cfs_time_seconds(timeoutl);
                                 lwi = LWI_TIMEOUT_INTERVAL(timeout,
                                                            cfs_time_seconds(1),
@@ -820,8 +847,7 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
                           remote_nb, npages, local_nb, oti, rc);
 
         if (rc == 0) {
-                repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
-                                         sizeof(*repbody));
+                repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
                 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
                 ost_drop_id(exp, &repbody->oa);
         }
@@ -869,7 +895,6 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
         struct lustre_handle     lockh = {0};
         struct lustre_capa      *capa = NULL;
         __u32                   *rcs;
-        __u32 size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
         int objcount, niocount, npages;
         int rc, i, j;
         obd_count                client_cksum = 0, server_cksum = 0;
@@ -890,10 +915,10 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 
         /* Check if there is eviction in progress, and if so, wait for it to
          * finish */
-        if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
+        if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
                 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
-                        !atomic_read(&exp->exp_obd->obd_evict_inprogress),
+                        !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
                         &lwi);
         }
         if (exp->exp_failed)
@@ -901,35 +926,47 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
 
         /* ost_body, ioobj & noibuf_remote are verified and swabbed in
          * ost_rw_hpreq_check(). */
-        body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
-        LASSERT(body != NULL);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+        if (body == NULL)
+                GOTO(out, rc = -EFAULT);
 
         if ((body->oa.o_flags & OBD_BRW_MEMALLOC) &&
             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
                 libcfs_memory_pressure_set();
 
-        objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
-                sizeof(*ioo);
-        ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
-                             objcount * sizeof(*ioo));
-        LASSERT(ioo != NULL);
+        objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
+                                        RCL_CLIENT) / sizeof(*ioo);
+        ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
+        if (ioo == NULL)
+                GOTO(out, rc = -EFAULT);
         for (niocount = i = 0; i < objcount; i++)
                 niocount += ioo[i].ioo_bufcnt;
 
-        remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
-                                   niocount * sizeof(*remote_nb));
-        LASSERT(remote_nb != NULL);
+        /*
+         * It'd be nice to have a capsule function to indicate how many elements
+         * there were in a buffer for an RMF that's declared to be an array.
+         * It's easy enough to compute the number of elements here though.
+         */
+        remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
+        if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
+            &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
+                GOTO(out, rc = -EFAULT);
 
-        if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
-                capa = lustre_unpack_incoming_capa(req, REQ_REC_OFF + 3);
+        if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+                capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
+                if (capa == NULL) {
+                        CERROR("Missing capability for OST BRW WRITE");
+                        GOTO(out, rc = -EFAULT);
+                }
+        }
 
-        size[REPLY_REC_OFF + 1] = niocount * sizeof(*rcs);
-        rc = lustre_pack_reply(req, 3, size, NULL);
+        req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
+                             niocount * sizeof(*rcs));
+        rc = req_capsule_server_pack(&req->rq_pill);
         if (rc != 0)
                 GOTO(out, rc);
         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
-        rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
-                             niocount * sizeof(*rcs));
+        rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
 
         /*
          * Per-thread array of struct niobuf_{local,remote}'s was allocated by
@@ -956,7 +993,10 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
                 GOTO(out_lock, rc = -ETIMEDOUT);
         }
 
-        ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa,  LCK_PW);
+        if (!lustre_handle_is_used(&lockh))
+                /* no needs to try to prolong lock if server is asked
+                 * to handle locking (= OBD_BRW_SRVLOCK) */
+                ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa,  LCK_PW);
 
         /* obd_preprw clobbers oa->valid, so save what we need */
         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
@@ -1011,7 +1051,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
                 do {
                         long timeoutl = req->rq_deadline -
                                 cfs_time_current_sec();
-                        cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
+                        cfs_duration_t timeout = timeoutl <= 0 ?
                                 CFS_TICK : cfs_time_seconds(timeoutl);
                         lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
                                                    ost_bulk_timeout, desc);
@@ -1053,8 +1093,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
         }
         no_reply = rc != 0;
 
-        repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
-                                 sizeof(*repbody));
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 
         if (unlikely(client_cksum != 0 && rc == 0)) {
@@ -1079,6 +1118,12 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
         /* Must commit after prep above in all cases */
         rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
                           remote_nb, npages, local_nb, oti, rc);
+        if (rc == -ENOTCONN)
+                /* quota acquire process has been given up because
+                 * either the client has been evicted or the client
+                 * has timed out the request already */
+                no_reply = 1;
+
         if (exp_connect_rmtclient(exp)) {
                 repbody->oa.o_uid = o_uid;
                 repbody->oa.o_gid = o_gid;
@@ -1176,47 +1221,64 @@ out:
         RETURN(rc);
 }
 
+/**
+ * Implementation of OST_SET_INFO.
+ *
+ * OST_SET_INFO is like ioctl(): heavily overloaded.  Specifically, it takes a
+ * "key" and a value RPC buffers as arguments, with the value's contents
+ * interpreted according to the key.
+ *
+ * Value types that need swabbing have swabbing done explicitly, either here or
+ * in functions called from here.  This should be corrected: all swabbing should
+ * be done in the capsule abstraction, as that will then allow us to move
+ * swabbing exclusively to the client without having to modify server code
+ * outside the capsule abstraction's implementation itself.  To correct this
+ * will require minor changes to the capsule abstraction; see the comments for
+ * req_capsule_extend() in layout.c.
+ */
 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
 {
         struct ost_body *body = NULL, *repbody;
-        __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
         char *key, *val = NULL;
         int keylen, vallen, rc = 0;
+        int is_grant_shrink = 0;
         ENTRY;
 
-        key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
+        key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
         if (key == NULL) {
                 DEBUG_REQ(D_HA, req, "no set_info key");
                 RETURN(-EFAULT);
         }
-        keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
+        keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
+                                      RCL_CLIENT);
 
-        if (KEY_IS(KEY_GRANT_SHRINK)) {
-                rc = lustre_pack_reply(req, 2, size, NULL);
-                if (rc)
-                        RETURN(rc);
-        } else {
-                rc = lustre_pack_reply(req, 1, NULL, NULL);
-                if (rc)
-                        RETURN(rc);
-        }
+        vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
+                                      RCL_CLIENT);
+
+        if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
+                /* In this case the value is actually an RMF_OST_BODY, so we
+                 * transmutate the type of this PTLRPC */
+                req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
+
+        rc = req_capsule_server_pack(&req->rq_pill);
+        if (rc)
+                RETURN(rc);
 
-        vallen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1);
         if (vallen) {
-                if (KEY_IS(KEY_GRANT_SHRINK)) {
-                        body = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
-                                                  sizeof(*body),
-                                                  lustre_swab_ost_body);
+                if (is_grant_shrink) {
+                        body = req_capsule_client_get(&req->rq_pill,
+                                                      &RMF_OST_BODY);
                         if (!body)
                                 RETURN(-EFAULT);
 
-                        repbody = lustre_msg_buf(req->rq_repmsg,
-                                                 REPLY_REC_OFF,
-                                                 sizeof(*repbody));
+                        repbody = req_capsule_server_get(&req->rq_pill,
+                                                         &RMF_OST_BODY);
                         memcpy(repbody, body, sizeof(*body));
                         val = (char*)repbody;
-                } else
-                        val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,0);
+                } else {
+                        val = req_capsule_client_get(&req->rq_pill,
+                                                     &RMF_SETINFO_VAL);
+                }
         }
 
         if (KEY_IS(KEY_EVICT_BY_NID)) {
@@ -1224,10 +1286,13 @@ static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
                         obd_export_evict_by_nid(exp->exp_obd, val);
                 GOTO(out, rc = 0);
         } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
-                /* Val's are not swabbed automatically */
+                if (vallen < sizeof(__u32))
+                        RETURN(-EFAULT);
                 __swab32s((__u32 *)val);
         }
 
+        /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
+         * a struct ost_body * value */
         rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
 out:
         lustre_msg_set_status(req->rq_repmsg, 0);
@@ -1241,8 +1306,6 @@ static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
         struct req_capsule *pill = &req->rq_pill;
         ENTRY;
 
-        req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
-
         /* this common part for get_info rpc */
         key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
         if (key == NULL) {
@@ -1346,7 +1409,7 @@ static int ost_llog_handle_connect(struct obd_export *exp,
         int rc;
         ENTRY;
 
-        body = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*body));
+        body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
         rc = obd_llog_connect(exp, body);
         RETURN(rc);
 }
@@ -1356,9 +1419,9 @@ do {                                                                    \
         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |          \
                                       OBD_CONNECT_RMT_CLIENT_FORCE |    \
                                       OBD_CONNECT_OSS_CAPA);            \
-        spin_lock(&exp->exp_lock);                                      \
+        cfs_spin_lock(&exp->exp_lock);                                  \
         exp->exp_connect_flags = reply->ocd_connect_flags;              \
-        spin_unlock(&exp->exp_lock);                                    \
+        cfs_spin_unlock(&exp->exp_lock);                                \
 } while (0)
 
 static int ost_init_sec_level(struct ptlrpc_request *req)
@@ -1455,9 +1518,9 @@ static int ost_init_sec_level(struct ptlrpc_request *req)
                         if (!filter->fo_fl_oss_capa)
                                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 
-                        spin_lock(&exp->exp_lock);
+                        cfs_spin_lock(&exp->exp_lock);
                         exp->exp_connect_flags = reply->ocd_connect_flags;
-                        spin_unlock(&exp->exp_lock);
+                        cfs_spin_unlock(&exp->exp_lock);
                 }
                 break;
         default:
@@ -1490,14 +1553,14 @@ static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
         }
 
         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
-                read_lock(&filter->fo_sptlrpc_lock);
+                cfs_read_lock(&filter->fo_sptlrpc_lock);
                 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
                                              req->rq_sp_from,
                                              req->rq_peer.nid,
                                              &flvr);
-                read_unlock(&filter->fo_sptlrpc_lock);
+                cfs_read_unlock(&filter->fo_sptlrpc_lock);
 
-                spin_lock(&exp->exp_lock);
+                cfs_spin_lock(&exp->exp_lock);
 
                 exp->exp_sp_peer = req->rq_sp_from;
                 exp->exp_flvr = flvr;
@@ -1511,7 +1574,7 @@ static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
                         rc = -EACCES;
                 }
 
-                spin_unlock(&exp->exp_lock);
+                cfs_spin_unlock(&exp->exp_lock);
         } else {
                 if (exp->exp_sp_peer != req->rq_sp_from) {
                         CERROR("RPC source %s doesn't match %s\n",
@@ -1623,6 +1686,10 @@ int ost_msg_check_version(struct lustre_msg *msg)
         return rc;
 }
 
+/**
+ * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
+ * not.
+ */
 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
                                    struct ldlm_lock *lock)
 {
@@ -1639,19 +1706,24 @@ static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
 
         /* As the request may be covered by several locks, do not look at
          * o_handle, look at the RPC IO region. */
-        body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
-                                  lustre_swab_obdo);
-        objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
-                sizeof(*ioo);
-        ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
-                             objcount * sizeof(*ioo));
-        LASSERT(ioo != NULL);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+        if (body == NULL)
+                RETURN(0);
+
+        objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
+                                        RCL_CLIENT) / sizeof(*ioo);
+        ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
+        if (ioo == NULL)
+                RETURN(0);
+
         for (niocount = i = 0; i < objcount; i++)
                 niocount += ioo[i].ioo_bufcnt;
 
-        nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
-                            niocount * sizeof(*nb));
-        LASSERT(nb != NULL);
+        nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
+        if (nb == NULL ||
+            niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
+            RCL_CLIENT) / sizeof(*nb)))
+                RETURN(0);
 
         mode = LCK_PW;
         if (opc == OST_READ)
@@ -1677,9 +1749,14 @@ static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
 }
 
 /**
- * Swab buffers needed to call ost_rw_prolong_locks() and call it.
- * Return the value from ost_rw_prolong_locks() which is non-zero if
- * there is a cancelled lock which is waiting for this IO request.
+ * High-priority queue request check for whether the given PTLRPC request (\a
+ * req) is blocking an LDLM lock cancel.
+ *
+ * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
+ * cancel, 0 if it is not, and -EFAULT if the request is malformed.
+ *
+ * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue.  This
+ * function looks only at OST_READs and OST_WRITEs.
  */
 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
 {
@@ -1693,21 +1770,25 @@ static int ost_rw_hpreq_check(struct ptlrpc_request *req)
         opc = lustre_msg_get_opc(req->rq_reqmsg);
         LASSERT(opc == OST_READ || opc == OST_WRITE);
 
-        body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
-        LASSERT(body != NULL);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+        if (body == NULL)
+                RETURN(-EFAULT);
 
-        objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
-                sizeof(*ioo);
-        ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
-                             objcount * sizeof(*ioo));
-        LASSERT(ioo != NULL);
+        objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
+                                        RCL_CLIENT) / sizeof(*ioo);
+        ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
+        if (ioo == NULL)
+                RETURN(-EFAULT);
 
         for (niocount = i = 0; i < objcount; i++)
                 niocount += ioo[i].ioo_bufcnt;
-        nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
-                            niocount * sizeof(*nb));
-        LASSERT(nb != NULL);
-        LASSERT(niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK));
+        nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
+        if (nb == NULL ||
+            niocount != (req_capsule_get_size(&req->rq_pill, &RMF_NIOBUF_REMOTE,
+            RCL_CLIENT) / sizeof(*nb)))
+                RETURN(-EFAULT);
+        if (niocount != 0 && (nb[0].flags & OBD_BRW_SRVLOCK))
+                RETURN(-EFAULT);
 
         mode = LCK_PW;
         if (opc == OST_READ)
@@ -1749,15 +1830,18 @@ static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa)
         RETURN(opd.opd_lock_match);
 }
 
+/**
+ * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
+ */
 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
                                       struct ldlm_lock *lock)
 {
         struct ost_body *body;
         ENTRY;
 
-        body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
-                                  lustre_swab_obdo);
-        LASSERT(body != NULL);
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+        if (body == NULL)
+                RETURN(0);  /* can't return -EFAULT here */
 
         if (body->oa.o_valid & OBD_MD_FLHANDLE &&
             body->oa.o_handle.cookie == lock->l_handle.h_cookie)
@@ -1765,13 +1849,19 @@ static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
         RETURN(0);
 }
 
+/**
+ * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
+ */
 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
 {
-        struct ost_body *body = lustre_msg_buf(req->rq_reqmsg,
-                                               REQ_REC_OFF, sizeof(*body));
-        LASSERT(body != NULL);
+        struct ost_body *body;
+
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+        if (body == NULL)
+                RETURN(-EFAULT);
+
         LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
-                !(body->oa.o_flags & OBD_FL_TRUNCLOCK));
+                !(body->oa.o_flags & OBD_FL_SRVLOCK));
 
         RETURN(ost_punch_prolong_locks(req, &body->oa));
 }
@@ -1798,18 +1888,32 @@ static int ost_hpreq_handler(struct ptlrpc_request *req)
                         struct niobuf_remote *nb;
                         struct obd_ioobj *ioo;
                         int objcount, niocount;
-                        int swab, i;
+                        int i;
 
-                        body = lustre_swab_reqbuf(req, REQ_REC_OFF,
-                                                  sizeof(*body),
-                                                  lustre_swab_obdo);
-                        if (!body) {
+                        /* RPCs on the H-P queue can be inspected before
+                         * ost_handler() initializes their pills, so we
+                         * initialize that here.  Capsule initialization is
+                         * idempotent, as is setting the pill's format (provided
+                         * it doesn't change).
+                         */
+                        req_capsule_init(&req->rq_pill, req, RCL_SERVER);
+                        if (opc == OST_READ)
+                                req_capsule_set(&req->rq_pill,
+                                                &RQF_OST_BRW_READ);
+                        else
+                                req_capsule_set(&req->rq_pill,
+                                                &RQF_OST_BRW_WRITE);
+
+                        body = req_capsule_client_get(&req->rq_pill,
+                                                      &RMF_OST_BODY);
+                        if (body == NULL) {
                                 CERROR("Missing/short ost_body\n");
                                 RETURN(-EFAULT);
                         }
-                        objcount = lustre_msg_buflen(req->rq_reqmsg,
-                                                     REQ_REC_OFF + 1) /
-                                sizeof(*ioo);
+                        objcount = req_capsule_get_size(&req->rq_pill,
+                                                        &RMF_OBD_IOOBJ,
+                                                        RCL_CLIENT) /
+                                                        sizeof(*ioo);
                         if (objcount == 0) {
                                 CERROR("Missing/short ioobj\n");
                                 RETURN(-EFAULT);
@@ -1819,18 +1923,14 @@ static int ost_hpreq_handler(struct ptlrpc_request *req)
                                 RETURN(-EFAULT);
                         }
 
-                        ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
-                                                 objcount * sizeof(*ioo),
-                                                 lustre_swab_obd_ioobj);
-                        if (!ioo) {
+                        ioo = req_capsule_client_get(&req->rq_pill,
+                                                     &RMF_OBD_IOOBJ);
+                        if (ioo == NULL) {
                                 CERROR("Missing/short ioobj\n");
                                 RETURN(-EFAULT);
                         }
 
-                        swab = ptlrpc_req_need_swab(req);
                         for (niocount = i = 0; i < objcount; i++) {
-                                if (i > 0 && swab)
-                                        lustre_swab_obd_ioobj(&ioo[i]);
                                 if (ioo[i].ioo_bufcnt == 0) {
                                         CERROR("ioo[%d] has zero bufcnt\n", i);
                                         RETURN(-EFAULT);
@@ -1838,39 +1938,34 @@ static int ost_hpreq_handler(struct ptlrpc_request *req)
                                 niocount += ioo[i].ioo_bufcnt;
                         }
                         if (niocount > PTLRPC_MAX_BRW_PAGES) {
-                                DEBUG_REQ(D_ERROR, req, "bulk has too many "
-                                          "pages (%d)", niocount);
+                                DEBUG_REQ(D_RPCTRACE, req,
+                                          "bulk has too many pages (%d)",
+                                          niocount);
                                 RETURN(-EFAULT);
                         }
 
-                        nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
-                                                niocount * sizeof(*nb),
-                                                lustre_swab_niobuf_remote);
-                        if (!nb) {
+                        nb = req_capsule_client_get(&req->rq_pill,
+                                                    &RMF_NIOBUF_REMOTE);
+                        if (nb == NULL) {
                                 CERROR("Missing/short niobuf\n");
                                 RETURN(-EFAULT);
                         }
 
-                        swab = ptlrpc_req_need_swab(req);
-                        if (swab) {
-                                /* swab remaining niobufs */
-                                for (i = 1; i < niocount; i++)
-                                        lustre_swab_niobuf_remote(&nb[i]);
-                        }
-
                         if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
                                 req->rq_ops = &ost_hpreq_rw;
                 } else if (opc == OST_PUNCH) {
-                        body = lustre_swab_reqbuf(req, REQ_REC_OFF,
-                                                  sizeof(*body),
-                                                  lustre_swab_obdo);
-                        if (!body) {
+                        req_capsule_init(&req->rq_pill, req, RCL_SERVER);
+                        req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
+
+                        body = req_capsule_client_get(&req->rq_pill,
+                                                      &RMF_OST_BODY);
+                        if (body == NULL) {
                                 CERROR("Missing/short ost_body\n");
                                 RETURN(-EFAULT);
                         }
 
                         if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
-                            !(body->oa.o_flags & OBD_FL_TRUNCLOCK))
+                            !(body->oa.o_flags & OBD_FL_SRVLOCK))
                                 req->rq_ops = &ost_hpreq_punch;
                 }
         }
@@ -1898,11 +1993,10 @@ int ost_handle(struct ptlrpc_request *req)
 
         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 
-        /* XXX identical to MDS */
         if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
                 int recovering;
 
-                if (req->rq_export == NULL) {
+                if (!class_connected_export(req->rq_export)) {
                         CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
                                lustre_msg_get_opc(req->rq_reqmsg),
                                libcfs_id2str(req->rq_peer));
@@ -1913,9 +2007,9 @@ int ost_handle(struct ptlrpc_request *req)
                 obd = req->rq_export->exp_obd;
 
                 /* Check for aborted recovery. */
-                spin_lock_bh(&obd->obd_processing_task_lock);
+                cfs_spin_lock_bh(&obd->obd_processing_task_lock);
                 recovering = obd->obd_recovering;
-                spin_unlock_bh(&obd->obd_processing_task_lock);
+                cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
                 if (recovering) {
                         rc = ost_filter_recovery_request(req, obd,
                                                          &should_process);
@@ -1960,16 +2054,16 @@ int ost_handle(struct ptlrpc_request *req)
                 break;
         case OST_CREATE:
                 CDEBUG(D_INODE, "create\n");
+                req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
                         RETURN(0);
-                if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
-                        GOTO(out, rc = -ENOSPC);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
                         GOTO(out, rc = -EROFS);
                 rc = ost_create(req->rq_export, req, oti);
                 break;
         case OST_DESTROY:
                 CDEBUG(D_INODE, "destroy\n");
+                req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
                         RETURN(0);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
@@ -1978,17 +2072,20 @@ int ost_handle(struct ptlrpc_request *req)
                 break;
         case OST_GETATTR:
                 CDEBUG(D_INODE, "getattr\n");
+                req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
                         RETURN(0);
                 rc = ost_getattr(req->rq_export, req);
                 break;
         case OST_SETATTR:
                 CDEBUG(D_INODE, "setattr\n");
+                req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
                         RETURN(0);
                 rc = ost_setattr(req->rq_export, req, oti);
                 break;
         case OST_WRITE:
+                req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
                 CDEBUG(D_INODE, "write\n");
                 /* req->rq_request_portal would be nice, if it was set */
                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
@@ -2009,6 +2106,7 @@ int ost_handle(struct ptlrpc_request *req)
                 /* ost_brw_write sends its own replies */
                 RETURN(rc);
         case OST_READ:
+                req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
                 CDEBUG(D_INODE, "read\n");
                 /* req->rq_request_portal would be nice, if it was set */
                 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
@@ -2026,6 +2124,7 @@ int ost_handle(struct ptlrpc_request *req)
                 RETURN(rc);
         case OST_PUNCH:
                 CDEBUG(D_INODE, "punch\n");
+                req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
                         RETURN(0);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
@@ -2034,12 +2133,14 @@ int ost_handle(struct ptlrpc_request *req)
                 break;
         case OST_STATFS:
                 CDEBUG(D_INODE, "statfs\n");
+                req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
                         RETURN(0);
                 rc = ost_statfs(req);
                 break;
         case OST_SYNC:
                 CDEBUG(D_INODE, "sync\n");
+                req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
                 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
                         RETURN(0);
                 rc = ost_sync(req->rq_export, req);
@@ -2051,6 +2152,7 @@ int ost_handle(struct ptlrpc_request *req)
                 break;
         case OST_GET_INFO:
                 DEBUG_REQ(D_INODE, req, "get_info");
+                req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
                 rc = ost_get_info(req->rq_export, req);
                 break;
 #ifdef HAVE_QUOTA_SUPPORT
@@ -2212,14 +2314,14 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
         int rc;
         ENTRY;
 
-        rc = cleanup_group_info();
+        rc = cfs_cleanup_group_info();
         if (rc)
                 RETURN(rc);
 
         lprocfs_ost_init_vars(&lvars);
         lprocfs_obd_setup(obd, lvars.obd_vars);
 
-        sema_init(&ost->ost_health_sem, 1);
+        cfs_sema_init(&ost->ost_health_sem, 1);
 
         if (oss_num_threads) {
                 /* If oss_num_threads is set, it is the min and the max. */
@@ -2230,7 +2332,8 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
                 oss_max_threads = oss_min_threads = oss_num_threads;
         } else {
                 /* Base min threads on memory and cpus */
-                oss_min_threads = num_possible_cpus() * CFS_NUM_CACHEPAGES >>
+                oss_min_threads =
+                        cfs_num_possible_cpus() * CFS_NUM_CACHEPAGES >>
                         (27 - CFS_PAGE_SHIFT);
                 if (oss_min_threads < OSS_THREADS_MIN)
                         oss_min_threads = OSS_THREADS_MIN;
@@ -2332,20 +2435,20 @@ static int ost_cleanup(struct obd_device *obd)
 
         ping_evictor_stop();
 
-        spin_lock_bh(&obd->obd_processing_task_lock);
+        cfs_spin_lock_bh(&obd->obd_processing_task_lock);
         if (obd->obd_recovering) {
                 target_cancel_recovery_timer(obd);
                 obd->obd_recovering = 0;
         }
-        spin_unlock_bh(&obd->obd_processing_task_lock);
+        cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
 
-        down(&ost->ost_health_sem);
+        cfs_down(&ost->ost_health_sem);
         ptlrpc_unregister_service(ost->ost_service);
         ptlrpc_unregister_service(ost->ost_create_service);
         ptlrpc_unregister_service(ost->ost_io_service);
         ost->ost_service = NULL;
         ost->ost_create_service = NULL;
-        up(&ost->ost_health_sem);
+        cfs_up(&ost->ost_health_sem);
 
         lprocfs_obd_cleanup(obd);
 
@@ -2357,11 +2460,11 @@ static int ost_health_check(struct obd_device *obd)
         struct ost_obd *ost = &obd->u.ost;
         int rc = 0;
 
-        down(&ost->ost_health_sem);
+        cfs_down(&ost->ost_health_sem);
         rc |= ptlrpc_service_health_check(ost->ost_service);
         rc |= ptlrpc_service_health_check(ost->ost_create_service);
         rc |= ptlrpc_service_health_check(ost->ost_io_service);
-        up(&ost->ost_health_sem);
+        cfs_up(&ost->ost_health_sem);
 
         /*
          * health_check to return 0 on healthy