Whamcloud - gitweb
LU-8842 mdc: Make IT_OPEN take lookup bits lock
[fs/lustre-release.git] / lustre / mdc / mdc_locks.c
index 1866103..c7f5ae9 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2014, Intel Corporation.
+ * Copyright (c) 2011, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #define DEBUG_SUBSYSTEM S_MDC
 
 #include <linux/module.h>
+
 #include <obd.h>
 #include <obd_class.h>
 #include <lustre_dlm.h>
-#include <lustre_fid.h> /* fid_res_name_eq() */
+#include <lustre_fid.h>
 #include <lustre_intent.h>
 #include <lustre_mdc.h>
 #include <lustre_net.h>
 #include <lustre_req_layout.h>
+#include <lustre_swab.h>
+
 #include "mdc_internal.h"
 
 struct mdc_getattr_args {
-        struct obd_export           *ga_exp;
-        struct md_enqueue_info      *ga_minfo;
-        struct ldlm_enqueue_info    *ga_einfo;
+       struct obd_export               *ga_exp;
+       struct md_enqueue_info          *ga_minfo;
 };
 
 int it_open_error(int phase, struct lookup_intent *it)
 {
        if (it_disposition(it, DISP_OPEN_LEASE)) {
                if (phase >= DISP_OPEN_LEASE)
-                       return it->d.lustre.it_status;
+                       return it->it_status;
+               else
+                       return 0;
+       }
+       if (it_disposition(it, DISP_OPEN_OPEN)) {
+               if (phase >= DISP_OPEN_OPEN)
+                       return it->it_status;
                else
                        return 0;
        }
-        if (it_disposition(it, DISP_OPEN_OPEN)) {
-                if (phase >= DISP_OPEN_OPEN)
-                        return it->d.lustre.it_status;
-                else
-                        return 0;
-        }
 
-        if (it_disposition(it, DISP_OPEN_CREATE)) {
-                if (phase >= DISP_OPEN_CREATE)
-                        return it->d.lustre.it_status;
-                else
-                        return 0;
-        }
+       if (it_disposition(it, DISP_OPEN_CREATE)) {
+               if (phase >= DISP_OPEN_CREATE)
+                       return it->it_status;
+               else
+                       return 0;
+       }
 
-        if (it_disposition(it, DISP_LOOKUP_EXECD)) {
-                if (phase >= DISP_LOOKUP_EXECD)
-                        return it->d.lustre.it_status;
-                else
-                        return 0;
-        }
+       if (it_disposition(it, DISP_LOOKUP_EXECD)) {
+               if (phase >= DISP_LOOKUP_EXECD)
+                       return it->it_status;
+               else
+                       return 0;
+       }
+
+       if (it_disposition(it, DISP_IT_EXECD)) {
+               if (phase >= DISP_IT_EXECD)
+                       return it->it_status;
+               else
+                       return 0;
+       }
+
+       CERROR("it disp: %X, status: %d\n", it->it_disposition, it->it_status);
+       LBUG();
 
-        if (it_disposition(it, DISP_IT_EXECD)) {
-                if (phase >= DISP_IT_EXECD)
-                        return it->d.lustre.it_status;
-                else
-                        return 0;
-        }
-        CERROR("it disp: %X, status: %d\n", it->d.lustre.it_disposition,
-               it->d.lustre.it_status);
-        LBUG();
         return 0;
 }
 EXPORT_SYMBOL(it_open_error);
 
 /* this must be called on a lockh that is known to have a referenced lock */
-int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
-                      __u64 *bits)
+int mdc_set_lock_data(struct obd_export *exp, const struct lustre_handle *lockh,
+                     void *data, __u64 *bits)
 {
        struct ldlm_lock *lock;
        struct inode *new_inode = data;
@@ -106,10 +105,10 @@ int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
         if(bits)
                 *bits = 0;
 
-        if (!*lockh)
-                RETURN(0);
+       if (!lustre_handle_is_used(lockh))
+               RETURN(0);
 
-        lock = ldlm_handle2lock((struct lustre_handle *)lockh);
+       lock = ldlm_handle2lock(lockh);
 
         LASSERT(lock != NULL);
         lock_res_and_lock(lock);
@@ -133,13 +132,13 @@ int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
         RETURN(0);
 }
 
-ldlm_mode_t mdc_lock_match(struct obd_export *exp, __u64 flags,
-                          const struct lu_fid *fid, ldlm_type_t type,
-                          ldlm_policy_data_t *policy, ldlm_mode_t mode,
-                          struct lustre_handle *lockh)
+enum ldlm_mode mdc_lock_match(struct obd_export *exp, __u64 flags,
+                             const struct lu_fid *fid, enum ldlm_type type,
+                             union ldlm_policy_data *policy,
+                             enum ldlm_mode mode, struct lustre_handle *lockh)
 {
        struct ldlm_res_id res_id;
-       ldlm_mode_t rc;
+       enum ldlm_mode rc;
        ENTRY;
 
        fid_build_reg_res_name(fid, &res_id);
@@ -150,23 +149,20 @@ ldlm_mode_t mdc_lock_match(struct obd_export *exp, __u64 flags,
        RETURN(rc);
 }
 
-int mdc_cancel_unused(struct obd_export *exp,
-                      const struct lu_fid *fid,
-                      ldlm_policy_data_t *policy,
-                      ldlm_mode_t mode,
-                      ldlm_cancel_flags_t flags,
-                      void *opaque)
+int mdc_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
+                     union ldlm_policy_data *policy, enum ldlm_mode mode,
+                     enum ldlm_cancel_flags flags, void *opaque)
 {
-        struct ldlm_res_id res_id;
-        struct obd_device *obd = class_exp2obd(exp);
-        int rc;
+       struct obd_device *obd = class_exp2obd(exp);
+       struct ldlm_res_id res_id;
+       int rc;
 
-        ENTRY;
+       ENTRY;
 
-        fid_build_reg_res_name(fid, &res_id);
-        rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
-                                             policy, mode, flags, opaque);
-        RETURN(rc);
+       fid_build_reg_res_name(fid, &res_id);
+       rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
+                                            policy, mode, flags, opaque);
+       RETURN(rc);
 }
 
 int mdc_null_inode(struct obd_export *exp,
@@ -193,28 +189,6 @@ int mdc_null_inode(struct obd_export *exp,
        RETURN(0);
 }
 
-/* find any ldlm lock of the inode in mdc
- * return 0    not find
- *        1    find one
- *      < 0    error */
-int mdc_find_cbdata(struct obd_export *exp,
-                    const struct lu_fid *fid,
-                    ldlm_iterator_t it, void *data)
-{
-        struct ldlm_res_id res_id;
-        int rc = 0;
-        ENTRY;
-
-        fid_build_reg_res_name((struct lu_fid*)fid, &res_id);
-        rc = ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace, &res_id,
-                                   it, data);
-        if (rc == LDLM_ITER_STOP)
-                RETURN(1);
-        else if (rc == LDLM_ITER_CONTINUE)
-                RETURN(0);
-        RETURN(rc);
-}
-
 static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
 {
        /* Don't hold error requests for replay. */
@@ -267,14 +241,14 @@ mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
        __u32                    lmmsize = op_data->op_data_size;
        struct list_head         cancels = LIST_HEAD_INIT(cancels);
        int                      count = 0;
-       int                      mode;
+       enum ldlm_mode           mode;
        int                      rc;
-        ENTRY;
+       ENTRY;
 
-        it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
+       it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
 
-        /* XXX: openlock is not cancelled for cross-refs. */
-        /* If inode is known, cancel conflicting OPEN locks. */
+       /* XXX: openlock is not cancelled for cross-refs. */
+       /* If inode is known, cancel conflicting OPEN locks. */
        if (fid_is_sane(&op_data->op_fid2)) {
                if (it->it_flags & MDS_OPEN_LEASE) { /* try to get lease */
                        if (it->it_flags & FMODE_WRITE)
@@ -317,6 +291,13 @@ mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
        req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
                             max(lmmsize, obddev->u.cli.cl_default_mds_easize));
 
+       req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
+                            RCL_CLIENT, op_data->op_file_secctx_name != NULL ?
+                            strlen(op_data->op_file_secctx_name) + 1 : 0);
+
+       req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX, RCL_CLIENT,
+                            op_data->op_file_secctx_size);
+
        rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
        if (rc < 0) {
                ptlrpc_request_free(req);
@@ -337,11 +318,6 @@ mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
 
        req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
                             obddev->u.cli.cl_max_mds_easize);
-
-        /* for remote client, fetch remote perm for current user */
-        if (client_is_remote(exp))
-                req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
-                                     sizeof(struct mdt_remote_perm));
         ptlrpc_request_set_replen(req);
         return req;
 }
@@ -439,9 +415,7 @@ static struct ptlrpc_request *mdc_intent_getattr_pack(struct obd_export *exp,
        struct obd_device       *obddev = class_exp2obd(exp);
        u64                      valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE |
                                         OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA |
-                                        OBD_MD_MEA |
-                                        (client_is_remote(exp) ?
-                                         OBD_MD_FLRMTPERM : OBD_MD_FLACL);
+                                        OBD_MD_MEA | OBD_MD_FLACL;
        struct ldlm_intent      *lit;
        int                      rc;
        __u32                    easize;
@@ -474,9 +448,6 @@ static struct ptlrpc_request *mdc_intent_getattr_pack(struct obd_export *exp,
        mdc_getattr_pack(req, valid, it->it_flags, op_data, easize);
 
        req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, easize);
-       if (client_is_remote(exp))
-               req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
-                                    sizeof(struct mdt_remote_perm));
        ptlrpc_request_set_replen(req);
        RETURN(req);
 }
@@ -552,7 +523,6 @@ static int mdc_finish_enqueue(struct obd_export *exp,
        struct req_capsule  *pill = &req->rq_pill;
        struct ldlm_request *lockreq;
        struct ldlm_reply   *lockrep;
-       struct lustre_intent_data *intent = &it->d.lustre;
        struct ldlm_lock    *lock;
        void                *lvb_data = NULL;
        __u32                lvb_len = 0;
@@ -587,16 +557,16 @@ static int mdc_finish_enqueue(struct obd_export *exp,
        lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
        LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
 
-       intent->it_disposition = (int)lockrep->lock_policy_res1;
-       intent->it_status = (int)lockrep->lock_policy_res2;
-       intent->it_lock_mode = einfo->ei_mode;
-       intent->it_lock_handle = lockh->cookie;
-       intent->it_data = req;
+       it->it_disposition = (int)lockrep->lock_policy_res1;
+       it->it_status = (int)lockrep->lock_policy_res2;
+       it->it_lock_mode = einfo->ei_mode;
+       it->it_lock_handle = lockh->cookie;
+       it->it_request = req;
 
        /* Technically speaking rq_transno must already be zero if
         * it_status is in error, so the check is a bit redundant */
-       if ((!req->rq_transno || intent->it_status < 0) && req->rq_replay)
-               mdc_clear_replay_flag(req, intent->it_status);
+       if ((!req->rq_transno || it->it_status < 0) && req->rq_replay)
+               mdc_clear_replay_flag(req, it->it_status);
 
         /* If we're doing an IT_OPEN which did not result in an actual
          * successful open, then we need to remove the bit which saves
@@ -606,14 +576,14 @@ static int mdc_finish_enqueue(struct obd_export *exp,
          * function without doing so, and try to replay a failed create
          * (bug 3440) */
         if (it->it_op & IT_OPEN && req->rq_replay &&
-           (!it_disposition(it, DISP_OPEN_OPEN) ||intent->it_status != 0))
-               mdc_clear_replay_flag(req, intent->it_status);
+           (!it_disposition(it, DISP_OPEN_OPEN) || it->it_status != 0))
+               mdc_clear_replay_flag(req, it->it_status);
 
        DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
-                 it->it_op, intent->it_disposition, intent->it_status);
+                 it->it_op, it->it_disposition, it->it_status);
 
         /* We know what to expect, so we do any byte flipping required here */
-        if (it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR)) {
+       if (it_has_reply_body(it)) {
                 struct mdt_body *body;
 
                 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
@@ -682,16 +652,6 @@ static int mdc_finish_enqueue(struct obd_export *exp,
                                               body->mbo_eadatasize);
                        }
                }
-
-               if (body->mbo_valid & OBD_MD_FLRMTPERM) {
-                        struct mdt_remote_perm *perm;
-
-                        LASSERT(client_is_remote(exp));
-                        perm = req_capsule_server_swab_get(pill, &RMF_ACL,
-                                                lustre_swab_mdt_remote_perm);
-                        if (perm == NULL)
-                                RETURN(-EPROTO);
-                }
         } else if (it->it_op & IT_LAYOUT) {
                /* maybe the lock was granted right away and layout
                 * is packed into RMF_DLM_LVB of req */
@@ -714,7 +674,7 @@ static int mdc_finish_enqueue(struct obd_export *exp,
            !(lockrep->lock_flags & LDLM_FL_BLOCKED_MASK)) {
                void *lmm;
 
-               LDLM_DEBUG(lock, "layout lock returned by: %s, lvb_len: %d\n",
+               LDLM_DEBUG(lock, "layout lock returned by: %s, lvb_len: %d",
                        ldlm_it2str(it->it_op), lvb_len);
 
                OBD_ALLOC_LARGE(lmm, lvb_len);
@@ -744,39 +704,41 @@ static int mdc_finish_enqueue(struct obd_export *exp,
 
 /* We always reserve enough space in the reply packet for a stripe MD, because
  * we don't know in advance the file type. */
-int mdc_enqueue(struct obd_export *exp,
-               struct ldlm_enqueue_info *einfo,
-               const union ldlm_policy_data *policy,
-               struct lookup_intent *it, struct md_op_data *op_data,
-               struct lustre_handle *lockh, __u64 extra_lock_flags)
+static int mdc_enqueue_base(struct obd_export *exp,
+                           struct ldlm_enqueue_info *einfo,
+                           const union ldlm_policy_data *policy,
+                           struct lookup_intent *it,
+                           struct md_op_data *op_data,
+                           struct lustre_handle *lockh,
+                           __u64 extra_lock_flags)
 {
-        struct obd_device     *obddev = class_exp2obd(exp);
-        struct ptlrpc_request *req = NULL;
-       __u64                  flags, saved_flags = extra_lock_flags;
-        int                    rc;
-        struct ldlm_res_id res_id;
-        static const ldlm_policy_data_t lookup_policy =
-                            { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
-        static const ldlm_policy_data_t update_policy =
-                            { .l_inodebits = { MDS_INODELOCK_UPDATE } };
-       static const ldlm_policy_data_t layout_policy =
-                           { .l_inodebits = { MDS_INODELOCK_LAYOUT } };
-       static const ldlm_policy_data_t getxattr_policy = {
-                             .l_inodebits = { MDS_INODELOCK_XATTR } };
-        int                    generation, resends = 0;
-        struct ldlm_reply     *lockrep;
-       enum lvb_type          lvb_type = 0;
-        ENTRY;
+       struct obd_device *obddev = class_exp2obd(exp);
+       struct ptlrpc_request *req = NULL;
+       __u64 flags, saved_flags = extra_lock_flags;
+       struct ldlm_res_id res_id;
+       static const union ldlm_policy_data lookup_policy = {
+                                 .l_inodebits = { MDS_INODELOCK_LOOKUP } };
+       static const union ldlm_policy_data update_policy = {
+                                 .l_inodebits = { MDS_INODELOCK_UPDATE } };
+       static const union ldlm_policy_data layout_policy = {
+                                 .l_inodebits = { MDS_INODELOCK_LAYOUT } };
+       static const union ldlm_policy_data getxattr_policy = {
+                                 .l_inodebits = { MDS_INODELOCK_XATTR } };
+       int generation, resends = 0;
+       struct ldlm_reply *lockrep;
+       enum lvb_type lvb_type = 0;
+       int rc;
+       ENTRY;
 
-        LASSERTF(!it || einfo->ei_type == LDLM_IBITS, "lock type %d\n",
-                 einfo->ei_type);
-        fid_build_reg_res_name(&op_data->op_fid1, &res_id);
+       LASSERTF(!it || einfo->ei_type == LDLM_IBITS, "lock type %d\n",
+                einfo->ei_type);
+       fid_build_reg_res_name(&op_data->op_fid1, &res_id);
 
        if (it != NULL) {
                LASSERT(policy == NULL);
 
                saved_flags |= LDLM_FL_HAS_INTENT;
-               if (it->it_op & (IT_OPEN | IT_UNLINK | IT_GETATTR | IT_READDIR))
+               if (it->it_op & (IT_UNLINK | IT_GETATTR | IT_READDIR))
                        policy = &update_policy;
                else if (it->it_op & IT_LAYOUT)
                        policy = &layout_policy;
@@ -817,25 +779,21 @@ resend:
         if (IS_ERR(req))
                 RETURN(PTR_ERR(req));
 
-       if (req != NULL && it && it->it_op & IT_CREAT)
-               /* ask ptlrpc not to resend on EINPROGRESS since we have our own
-                * retry logic */
-               req->rq_no_retry_einprogress = 1;
-
         if (resends) {
                 req->rq_generation_set = 1;
                 req->rq_import_generation = generation;
                 req->rq_sent = cfs_time_current_sec() + resends;
         }
 
-        /* It is important to obtain rpc_lock first (if applicable), so that
-         * threads that are serialised with rpc_lock are not polluting our
-         * rpcs in flight counter. We do not do flock request limiting, though*/
-        if (it) {
-                mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+       /* It is important to obtain modify RPC slot first (if applicable), so
+        * that threads that are waiting for a modify RPC slot are not polluting
+        * our rpcs in flight counter.
+        * We do not do flock request limiting, though */
+       if (it) {
+               mdc_get_mod_rpc_slot(req, it);
                rc = obd_get_request_slot(&obddev->u.cli);
-                if (rc != 0) {
-                        mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+               if (rc != 0) {
+                       mdc_put_mod_rpc_slot(req, it);
                         mdc_clear_replay_flag(req, 0);
                         ptlrpc_req_finished(req);
                         RETURN(rc);
@@ -861,7 +819,7 @@ resend:
        }
 
        obd_put_request_slot(&obddev->u.cli);
-       mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+       mdc_put_mod_rpc_slot(req, it);
 
        if (rc < 0) {
                CDEBUG(D_INFO, "%s: ldlm_cli_enqueue failed: rc = %d\n",
@@ -878,10 +836,10 @@ resend:
        lockrep->lock_policy_res2 =
                ptlrpc_status_ntoh(lockrep->lock_policy_res2);
 
-        /* Retry the create infinitely when we get -EINPROGRESS from
-         * server. This is required by the new quota design. */
-        if (it && it->it_op & IT_CREAT &&
-            (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
+       /* Retry infinitely when the server returns -EINPROGRESS for the
+        * intent operation, when server returns -EINPROGRESS for acquiring
+        * intent lock, we'll retry in after_reply(). */
+       if (it && (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
                 mdc_clear_replay_flag(req, rc);
                 ptlrpc_req_finished(req);
                 resends++;
@@ -906,14 +864,23 @@ resend:
                }
                ptlrpc_req_finished(req);
 
-               it->d.lustre.it_lock_handle = 0;
-               it->d.lustre.it_lock_mode = 0;
-               it->d.lustre.it_data = NULL;
+               it->it_lock_handle = 0;
+               it->it_lock_mode = 0;
+               it->it_request = NULL;
        }
 
        RETURN(rc);
 }
 
+int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
+               const union ldlm_policy_data *policy,
+               struct md_op_data *op_data,
+               struct lustre_handle *lockh, __u64 extra_lock_flags)
+{
+       return mdc_enqueue_base(exp, einfo, policy, NULL,
+                               op_data, lockh, extra_lock_flags);
+}
+
 static int mdc_finish_intent_lock(struct obd_export *exp,
                                   struct ptlrpc_request *request,
                                   struct md_op_data *op_data,
@@ -921,9 +888,8 @@ static int mdc_finish_intent_lock(struct obd_export *exp,
                                   struct lustre_handle *lockh)
 {
         struct lustre_handle old_lock;
-        struct mdt_body *mdt_body;
         struct ldlm_lock *lock;
-        int rc;
+       int rc = 0;
        ENTRY;
 
         LASSERT(request != NULL);
@@ -933,96 +899,117 @@ static int mdc_finish_intent_lock(struct obd_export *exp,
        if (it->it_op & IT_READDIR)
                RETURN(0);
 
-        if (!it_disposition(it, DISP_IT_EXECD)) {
-                /* The server failed before it even started executing the
-                 * intent, i.e. because it couldn't unpack the request. */
-                LASSERT(it->d.lustre.it_status != 0);
-                RETURN(it->d.lustre.it_status);
-        }
-        rc = it_open_error(DISP_IT_EXECD, it);
-        if (rc)
-                RETURN(rc);
-
-        mdt_body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
-        LASSERT(mdt_body != NULL);      /* mdc_enqueue checked */
-
-        rc = it_open_error(DISP_LOOKUP_EXECD, it);
-        if (rc)
-                RETURN(rc);
-
-        /* keep requests around for the multiple phases of the call
-         * this shows the DISP_XX must guarantee we make it into the call
-         */
-        if (!it_disposition(it, DISP_ENQ_CREATE_REF) &&
-            it_disposition(it, DISP_OPEN_CREATE) &&
-            !it_open_error(DISP_OPEN_CREATE, it)) {
-                it_set_disposition(it, DISP_ENQ_CREATE_REF);
-                ptlrpc_request_addref(request); /* balanced in ll_create_node */
-        }
-        if (!it_disposition(it, DISP_ENQ_OPEN_REF) &&
-            it_disposition(it, DISP_OPEN_OPEN) &&
-            !it_open_error(DISP_OPEN_OPEN, it)) {
-                it_set_disposition(it, DISP_ENQ_OPEN_REF);
-                ptlrpc_request_addref(request); /* balanced in ll_file_open */
-                /* BUG 11546 - eviction in the middle of open rpc processing */
-                OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_ENQUEUE_PAUSE, obd_timeout);
-        }
+       if (it->it_op & (IT_GETXATTR | IT_LAYOUT)) {
+               if (it->it_status != 0)
+                       GOTO(out, rc = it->it_status);
+       } else {
+               if (!it_disposition(it, DISP_IT_EXECD)) {
+                       /* The server failed before it even started executing
+                        * the intent, i.e. because it couldn't unpack the
+                        * request.
+                        */
+                       LASSERT(it->it_status != 0);
+                       GOTO(out, rc = it->it_status);
+               }
+               rc = it_open_error(DISP_IT_EXECD, it);
+               if (rc)
+                       GOTO(out, rc);
+
+               rc = it_open_error(DISP_LOOKUP_EXECD, it);
+               if (rc)
+                       GOTO(out, rc);
+
+               /* keep requests around for the multiple phases of the call
+                * this shows the DISP_XX must guarantee we make it into the
+                * call
+                */
+               if (!it_disposition(it, DISP_ENQ_CREATE_REF) &&
+                   it_disposition(it, DISP_OPEN_CREATE) &&
+                   !it_open_error(DISP_OPEN_CREATE, it)) {
+                       it_set_disposition(it, DISP_ENQ_CREATE_REF);
+                       /* balanced in ll_create_node */
+                       ptlrpc_request_addref(request);
+               }
+               if (!it_disposition(it, DISP_ENQ_OPEN_REF) &&
+                   it_disposition(it, DISP_OPEN_OPEN) &&
+                   !it_open_error(DISP_OPEN_OPEN, it)) {
+                       it_set_disposition(it, DISP_ENQ_OPEN_REF);
+                       /* balanced in ll_file_open */
+                       ptlrpc_request_addref(request);
+                       /* BUG 11546 - eviction in the middle of open rpc
+                        * processing
+                        */
+                       OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_ENQUEUE_PAUSE,
+                                        obd_timeout);
+               }
 
-        if (it->it_op & IT_CREAT) {
-                /* XXX this belongs in ll_create_it */
-        } else if (it->it_op == IT_OPEN) {
-                LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
-        } else {
-                LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_LAYOUT));
-        }
+               if (it->it_op & IT_CREAT) {
+                       /* XXX this belongs in ll_create_it */
+               } else if (it->it_op == IT_OPEN) {
+                       LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
+               } else {
+                       LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
+               }
+       }
 
-        /* If we already have a matching lock, then cancel the new
-         * one.  We have to set the data here instead of in
-         * mdc_enqueue, because we need to use the child's inode as
-         * the l_ast_data to match, and that's not available until
-         * intent_finish has performed the iget().) */
-        lock = ldlm_handle2lock(lockh);
-        if (lock) {
-                ldlm_policy_data_t policy = lock->l_policy_data;
-                LDLM_DEBUG(lock, "matching against this");
-
-               LASSERTF(fid_res_name_eq(&mdt_body->mbo_fid1,
-                                        &lock->l_resource->lr_name),
-                        "Lock res_id: "DLDLMRES", fid: "DFID"\n",
-                        PLDLMRES(lock->l_resource), PFID(&mdt_body->mbo_fid1));
+       /* If we already have a matching lock, then cancel the new
+        * one.  We have to set the data here instead of in
+        * mdc_enqueue, because we need to use the child's inode as
+        * the l_ast_data to match, and that's not available until
+        * intent_finish has performed the iget().) */
+       lock = ldlm_handle2lock(lockh);
+       if (lock) {
+               union ldlm_policy_data policy = lock->l_policy_data;
+               LDLM_DEBUG(lock, "matching against this");
+
+               if (it_has_reply_body(it)) {
+                       struct mdt_body *body;
+
+                       body = req_capsule_server_get(&request->rq_pill,
+                                                     &RMF_MDT_BODY);
+                       /* mdc_enqueue checked */
+                       LASSERT(body != NULL);
+                       LASSERTF(fid_res_name_eq(&body->mbo_fid1,
+                                                &lock->l_resource->lr_name),
+                                "Lock res_id: "DLDLMRES", fid: "DFID"\n",
+                                PLDLMRES(lock->l_resource),
+                                PFID(&body->mbo_fid1));
+               }
                LDLM_LOCK_PUT(lock);
 
                 memcpy(&old_lock, lockh, sizeof(*lockh));
                 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
                                     LDLM_IBITS, &policy, LCK_NL, &old_lock, 0)) {
-                        ldlm_lock_decref_and_cancel(lockh,
-                                                    it->d.lustre.it_lock_mode);
-                        memcpy(lockh, &old_lock, sizeof(old_lock));
-                        it->d.lustre.it_lock_handle = lockh->cookie;
-                }
-        }
+                       ldlm_lock_decref_and_cancel(lockh, it->it_lock_mode);
+                       memcpy(lockh, &old_lock, sizeof(old_lock));
+                       it->it_lock_handle = lockh->cookie;
+               }
+       }
+
+       EXIT;
+out:
        CDEBUG(D_DENTRY,"D_IT dentry %.*s intent: %s status %d disp %x rc %d\n",
                (int)op_data->op_namelen, op_data->op_name,
-               ldlm_it2str(it->it_op), it->d.lustre.it_status,
-               it->d.lustre.it_disposition, rc);
-       RETURN(rc);
+               ldlm_it2str(it->it_op), it->it_status,
+               it->it_disposition, rc);
+       return rc;
 }
 
 int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
-                        struct lu_fid *fid, __u64 *bits)
+                       struct lu_fid *fid, __u64 *bits)
 {
-        /* We could just return 1 immediately, but since we should only
-         * be called in revalidate_it if we already have a lock, let's
-         * verify that. */
-        struct ldlm_res_id res_id;
-        struct lustre_handle lockh;
-        ldlm_policy_data_t policy;
-        ldlm_mode_t mode;
-        ENTRY;
+       /* We could just return 1 immediately, but since we should only
+        * be called in revalidate_it if we already have a lock, let's
+        * verify that. */
+       struct ldlm_res_id res_id;
+       struct lustre_handle lockh;
+       union ldlm_policy_data policy;
+       enum ldlm_mode mode;
+       ENTRY;
 
-        if (it->d.lustre.it_lock_handle) {
-                lockh.cookie = it->d.lustre.it_lock_handle;
-                mode = ldlm_revalidate_lock_handle(&lockh, bits);
+       if (it->it_lock_handle) {
+               lockh.cookie = it->it_lock_handle;
+               mode = ldlm_revalidate_lock_handle(&lockh, bits);
         } else {
                 fid_build_reg_res_name(fid, &res_id);
                 switch (it->it_op) {
@@ -1062,15 +1049,15 @@ int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
                                      &lockh);
         }
 
-        if (mode) {
-                it->d.lustre.it_lock_handle = lockh.cookie;
-                it->d.lustre.it_lock_mode = mode;
-        } else {
-                it->d.lustre.it_lock_handle = 0;
-                it->d.lustre.it_lock_mode = 0;
-        }
+       if (mode) {
+               it->it_lock_handle = lockh.cookie;
+               it->it_lock_mode = mode;
+       } else {
+               it->it_lock_handle = 0;
+               it->it_lock_mode = 0;
+       }
 
-        RETURN(!!mode);
+       RETURN(!!mode);
 }
 
 /*
@@ -1089,15 +1076,15 @@ int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
  * ll_create/ll_open gets called.
  *
  * The server will return to us, in it_disposition, an indication of
- * exactly what d.lustre.it_status refers to.
+ * exactly what it_status refers to.
  *
- * If DISP_OPEN_OPEN is set, then d.lustre.it_status refers to the open() call,
+ * If DISP_OPEN_OPEN is set, then it_status refers to the open() call,
  * otherwise if DISP_OPEN_CREATE is set, then it status is the
  * creation failure mode.  In either case, one of DISP_LOOKUP_NEG or
  * DISP_LOOKUP_POS will be set, indicating whether the child lookup
  * was successful.
  *
- * Else, if DISP_LOOKUP_EXECD then d.lustre.it_status is the rc of the
+ * Else, if DISP_LOOKUP_EXECD then it_status is the rc of the
  * child lookup.
  */
 int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
@@ -1116,7 +1103,7 @@ int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
        LASSERT(it);
 
        CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
-               ", intent: %s flags %#"LPF64"o\n", (int)op_data->op_namelen,
+               ", intent: %s flags %#llo\n", (int)op_data->op_namelen,
                op_data->op_name, PFID(&op_data->op_fid2),
                PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
                it->it_flags);
@@ -1127,7 +1114,7 @@ int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
                /* We could just return 1 immediately, but since we should only
                 * be called in revalidate_it if we already have a lock, let's
                 * verify that. */
-               it->d.lustre.it_lock_handle = 0;
+               it->it_lock_handle = 0;
                rc = mdc_revalidate_lock(exp, it, &op_data->op_fid2, NULL);
                /* Only return failure if it was not GETATTR by cfid
                   (from inode_revalidate) */
@@ -1144,12 +1131,12 @@ int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
                }
        }
 
-       rc = mdc_enqueue(exp, &einfo, NULL, it, op_data, &lockh,
-                        extra_lock_flags);
+       rc = mdc_enqueue_base(exp, &einfo, NULL, it, op_data, &lockh,
+                             extra_lock_flags);
        if (rc < 0)
                RETURN(rc);
 
-       *reqp = it->d.lustre.it_data;
+       *reqp = it->it_request;
         rc = mdc_finish_intent_lock(exp, *reqp, op_data, it, &lockh);
         RETURN(rc);
 }
@@ -1158,16 +1145,16 @@ static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
                                               struct ptlrpc_request *req,
                                               void *args, int rc)
 {
-        struct mdc_getattr_args  *ga = args;
-        struct obd_export        *exp = ga->ga_exp;
-        struct md_enqueue_info   *minfo = ga->ga_minfo;
-        struct ldlm_enqueue_info *einfo = ga->ga_einfo;
-        struct lookup_intent     *it;
-        struct lustre_handle     *lockh;
-        struct obd_device        *obddev;
+       struct mdc_getattr_args  *ga = args;
+       struct obd_export        *exp = ga->ga_exp;
+       struct md_enqueue_info   *minfo = ga->ga_minfo;
+       struct ldlm_enqueue_info *einfo = &minfo->mi_einfo;
+       struct lookup_intent     *it;
+       struct lustre_handle     *lockh;
+       struct obd_device        *obddev;
        struct ldlm_reply        *lockrep;
        __u64                     flags = LDLM_FL_HAS_INTENT;
-        ENTRY;
+       ENTRY;
 
         it    = &minfo->mi_it;
         lockh = &minfo->mi_lockh;
@@ -1200,14 +1187,12 @@ static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
         EXIT;
 
 out:
-        OBD_FREE_PTR(einfo);
         minfo->mi_cb(req, minfo, rc);
         return 0;
 }
 
 int mdc_intent_getattr_async(struct obd_export *exp,
-                            struct md_enqueue_info *minfo,
-                            struct ldlm_enqueue_info *einfo)
+                            struct md_enqueue_info *minfo)
 {
        struct md_op_data       *op_data = &minfo->mi_data;
        struct lookup_intent    *it = &minfo->mi_it;
@@ -1215,19 +1200,14 @@ int mdc_intent_getattr_async(struct obd_export *exp,
        struct mdc_getattr_args *ga;
        struct obd_device       *obddev = class_exp2obd(exp);
        struct ldlm_res_id       res_id;
-       /*XXX: Both MDS_INODELOCK_LOOKUP and MDS_INODELOCK_UPDATE are needed
-        *     for statahead currently. Consider CMD in future, such two bits
-        *     maybe managed by different MDS, should be adjusted then. */
-       ldlm_policy_data_t       policy = {
-                                       .l_inodebits = { MDS_INODELOCK_LOOKUP |
-                                                        MDS_INODELOCK_UPDATE }
-                                };
+       union ldlm_policy_data policy = {
+                               .l_inodebits = { MDS_INODELOCK_LOOKUP |
+                                                MDS_INODELOCK_UPDATE } };
        int                      rc = 0;
        __u64                    flags = LDLM_FL_HAS_INTENT;
        ENTRY;
 
-       CDEBUG(D_DLMTRACE, "name: %.*s in inode "DFID", intent: %s flags %#"
-               LPF64"o\n",
+       CDEBUG(D_DLMTRACE, "name: %.*s in inode "DFID", intent: %s flags %#llo\n",
                (int)op_data->op_namelen, op_data->op_name,
                PFID(&op_data->op_fid1), ldlm_it2str(it->it_op), it->it_flags);
 
@@ -1237,27 +1217,26 @@ int mdc_intent_getattr_async(struct obd_export *exp,
                RETURN(PTR_ERR(req));
 
        rc = obd_get_request_slot(&obddev->u.cli);
-        if (rc != 0) {
-                ptlrpc_req_finished(req);
-                RETURN(rc);
-        }
+       if (rc != 0) {
+               ptlrpc_req_finished(req);
+               RETURN(rc);
+       }
 
-        rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
-                             0, LVB_T_NONE, &minfo->mi_lockh, 1);
-        if (rc < 0) {
+       rc = ldlm_cli_enqueue(exp, &req, &minfo->mi_einfo, &res_id, &policy,
+                             &flags, NULL, 0, LVB_T_NONE, &minfo->mi_lockh, 1);
+       if (rc < 0) {
                obd_put_request_slot(&obddev->u.cli);
-                ptlrpc_req_finished(req);
-                RETURN(rc);
-        }
+               ptlrpc_req_finished(req);
+               RETURN(rc);
+       }
 
-        CLASSERT(sizeof(*ga) <= sizeof(req->rq_async_args));
-        ga = ptlrpc_req_async_args(req);
-        ga->ga_exp = exp;
-        ga->ga_minfo = minfo;
-        ga->ga_einfo = einfo;
+       CLASSERT(sizeof(*ga) <= sizeof(req->rq_async_args));
+       ga = ptlrpc_req_async_args(req);
+       ga->ga_exp = exp;
+       ga->ga_minfo = minfo;
 
-        req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
-        ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
+       req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
+       ptlrpcd_add_req(req);
 
-        RETURN(0);
+       RETURN(0);
 }