Whamcloud - gitweb
land minor fixes from b_hd_sec:
[fs/lustre-release.git] / lustre / mdc / mdc_locks.c
index 7998565..3addfcf 100644 (file)
 #include <linux/obd_class.h>
 #include <linux/lustre_mds.h>
 #include <linux/lustre_dlm.h>
-//#include <linux/lustre_smfs.h>
-//#include <linux/lustre_snap.h>
+#include <linux/lustre_sec.h>
 #include <linux/lprocfs_status.h>
+#include <linux/lustre_acl.h>
+#include <linux/lustre_lite.h>
 #include "mdc_internal.h"
 
 int it_disposition(struct lookup_intent *it, int flag)
 {
-        return it->d.lustre.it_disposition & flag;
+        return LUSTRE_IT(it)->it_disposition & flag;
 }
 EXPORT_SYMBOL(it_disposition);
 
 void it_set_disposition(struct lookup_intent *it, int flag)
 {
-        it->d.lustre.it_disposition |= flag;
+        LUSTRE_IT(it)->it_disposition |= flag;
 }
 EXPORT_SYMBOL(it_set_disposition);
 
-static void mdc_fid2mdc_op_data(struct mdc_op_data *data,
-                                struct ll_fid *f1, struct ll_fid *f2,
-                                const char *name, int namelen, int mode)
+static void mdc_id2mdc_data(struct mdc_op_data *data,
+                            struct lustre_id *f1, 
+                            struct lustre_id *f2,
+                            const char *name, 
+                            int namelen, int mode)
 {
         LASSERT(data);
         LASSERT(f1);
 
-        data->fid1 = *f1;
+        data->id1 = *f1;
         if (f2)
-                data->fid2 = *f2;
-        else
-                memset(&data->fid2, 0, sizeof(data->fid2));
+                data->id2 = *f2;
+
+        data->valid = 0;
         data->name = name;
         data->namelen = namelen;
         data->create_mode = mode;
@@ -88,33 +91,33 @@ int it_open_error(int phase, struct lookup_intent *it)
 {
         if (it_disposition(it, DISP_OPEN_OPEN)) {
                 if (phase == DISP_OPEN_OPEN)
-                        return it->d.lustre.it_status;
+                        return LUSTRE_IT(it)->it_status;
                 else
                         return 0;
         }
 
         if (it_disposition(it, DISP_OPEN_CREATE)) {
                 if (phase == DISP_OPEN_CREATE)
-                        return it->d.lustre.it_status;
+                        return LUSTRE_IT(it)->it_status;
                 else
                         return 0;
         }
 
         if (it_disposition(it, DISP_LOOKUP_EXECD)) {
                 if (phase == DISP_LOOKUP_EXECD)
-                        return it->d.lustre.it_status;
+                        return LUSTRE_IT(it)->it_status;
                 else
                         return 0;
         }
 
         if (it_disposition(it, DISP_IT_EXECD)) {
                 if (phase == DISP_IT_EXECD)
-                        return it->d.lustre.it_status;
+                        return LUSTRE_IT(it)->it_status;
                 else
                         return 0;
         }
-        CERROR("it disp: %X, status: %d\n", it->d.lustre.it_disposition,
-               it->d.lustre.it_status);
+        CERROR("it disp: %X, status: %d\n", LUSTRE_IT(it)->it_disposition,
+               LUSTRE_IT(it)->it_status);
         LBUG();
         return 0;
 }
@@ -135,7 +138,7 @@ int mdc_set_lock_data(struct obd_export *exp, __u64 *l, void *data)
         lock = ldlm_handle2lock(lockh);
 
         LASSERT(lock != NULL);
-        l_lock(&lock->l_resource->lr_namespace->ns_lock);
+        lock_res_and_lock(lock);
 #ifdef __KERNEL__
         if (lock->l_ast_data && lock->l_ast_data != data) {
                 struct inode *new_inode = data;
@@ -144,12 +147,12 @@ int mdc_set_lock_data(struct obd_export *exp, __u64 *l, void *data)
                          "Found existing inode %p/%lu/%u state %lu in lock: "
                          "setting data to %p/%lu/%u\n", old_inode,
                          old_inode->i_ino, old_inode->i_generation,
-                         old_inode->i_state,
-                         new_inode, new_inode->i_ino, new_inode->i_generation);
+                         old_inode->i_state, new_inode, new_inode->i_ino,
+                         new_inode->i_generation);
         }
 #endif
         lock->l_ast_data = data;
-        l_unlock(&lock->l_resource->lr_namespace->ns_lock);
+        unlock_res_and_lock(lock);
         LDLM_LOCK_PUT(lock);
 
         EXIT;
@@ -157,22 +160,38 @@ int mdc_set_lock_data(struct obd_export *exp, __u64 *l, void *data)
 }
 EXPORT_SYMBOL(mdc_set_lock_data);
 
-int mdc_change_cbdata(struct obd_export *exp, struct ll_fid *fid, 
+int mdc_change_cbdata(struct obd_export *exp, struct lustre_id *id, 
                       ldlm_iterator_t it, void *data)
 {
         struct ldlm_res_id res_id = { .name = {0} };
         ENTRY;
 
-        res_id.name[0] = fid->id;
-        res_id.name[1] = fid->generation;
+        res_id.name[0] = id_fid(id);
+        res_id.name[1] = id_group(id);
 
-        ldlm_change_cbdata(class_exp2obd(exp)->obd_namespace, &res_id, it, 
-                           data);
+        ldlm_change_cbdata(class_exp2obd(exp)->obd_namespace,
+                           &res_id, it, data);
 
         EXIT;
         return 0;
 }
 
+static inline void
+mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
+{
+        /* Don't hold error requests for replay. */
+        if (req->rq_replay) {
+                unsigned long irqflags;
+                spin_lock_irqsave(&req->rq_lock, irqflags);
+                req->rq_replay = 0;
+                spin_unlock_irqrestore(&req->rq_lock, irqflags);
+        }
+        if (rc && req->rq_transno != 0) {
+                DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
+                LBUG();
+        }
+}
+
 /* 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,
@@ -188,22 +207,22 @@ int mdc_enqueue(struct obd_export *exp,
                 void *cb_data)
 {
         struct ptlrpc_request *req;
+        struct ldlm_res_id res_id = {
+                .name = {id_fid(&data->id1), id_group(&data->id1)}
+        };
         struct obd_device *obddev = class_exp2obd(exp);
-        struct ldlm_res_id res_id =
-                { .name = {data->fid1.id, data->fid1.generation} };
         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
         struct ldlm_intent *lit;
         struct ldlm_request *lockreq;
-        struct ldlm_reply *dlm_rep;
         int reqsize[6] = {[MDS_REQ_SECDESC_OFF] = 0,
                           [MDS_REQ_INTENT_LOCKREQ_OFF] = sizeof(*lockreq),
                           [MDS_REQ_INTENT_IT_OFF] = sizeof(*lit)};
-        int repsize[4] = {sizeof(struct ldlm_reply),
+        int repsize[5] = {sizeof(struct ldlm_reply),
                           sizeof(struct mds_body),
-                          obddev->u.cli.cl_max_mds_easize,
-                          obddev->u.cli.cl_max_mds_cookiesize};
+                          obddev->u.cli.cl_max_mds_easize};
         int req_buffers = 3, reply_buffers = 0;
         int rc, flags = LDLM_FL_HAS_INTENT;
+        struct ldlm_reply *dlm_rep = NULL;
         void *eadata;
         unsigned long irqflags;
         ENTRY;
@@ -211,7 +230,7 @@ int mdc_enqueue(struct obd_export *exp,
 //        LDLM_DEBUG_NOLOCK("mdsintent=%s,name=%s,dir=%lu",
 //                          ldlm_it2str(it->it_op), it_name, it_inode->i_ino);
 
-        reqsize[0] = mdc_get_secdesc_size();
+        reqsize[0] = lustre_secdesc_size();
 
         if (it->it_op & IT_OPEN) {
                 it->it_create_mode |= S_IFREG;
@@ -220,6 +239,7 @@ int mdc_enqueue(struct obd_export *exp,
                 reqsize[req_buffers++] = sizeof(struct mds_rec_create);
                 reqsize[req_buffers++] = data->namelen + 1;
                 reqsize[req_buffers++] = obddev->u.cli.cl_max_mds_easize;
+
                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
                                       LDLM_ENQUEUE, req_buffers, reqsize, NULL);
                 if (!req)
@@ -236,19 +256,25 @@ int mdc_enqueue(struct obd_export *exp,
 
                 /* pack the intended request */
                 mdc_open_pack(req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF, data,
-                              it->it_create_mode, 0, it->it_flags,
-                              lmm, lmmsize);
+                              it->it_create_mode, 0, it->it_flags, lmm, lmmsize);
                 /* get ready for the reply */
-                reply_buffers = 3;
-                req->rq_replen = lustre_msg_size(3, repsize);
+                repsize[3] = 4;
+                repsize[4] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
+                reply_buffers = 5;
+                req->rq_replen = lustre_msg_size(5, repsize);
         } else if (it->it_op & (IT_GETATTR | IT_LOOKUP | IT_CHDIR)) {
-                int valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
+                __u64 valid = data->valid | OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE |
+                              OBD_MD_FLACL;
+
+                /* we don't expect xattr retrieve could reach here */
+                LASSERT(!(valid & (OBD_MD_FLXATTR | OBD_MD_FLXATTRLIST)));
 
                 reqsize[req_buffers++] = sizeof(struct mds_body);
                 reqsize[req_buffers++] = data->namelen + 1;
 
                 if (it->it_op & IT_GETATTR)
                         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
+
                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
                                       LDLM_ENQUEUE, req_buffers, reqsize, NULL);
 
@@ -263,9 +289,12 @@ int mdc_enqueue(struct obd_export *exp,
                 /* pack the intended request */
                 mdc_getattr_pack(req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
                                  valid, it->it_flags, data);
+                
                 /* get ready for the reply */
-                reply_buffers = 3;
-                req->rq_replen = lustre_msg_size(3, repsize);
+                repsize[3] = 4;
+                repsize[4] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
+                reply_buffers = 5;
+                req->rq_replen = lustre_msg_size(5, repsize);
         } else if (it->it_op == IT_READDIR) {
                 policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
                 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
@@ -301,7 +330,7 @@ int mdc_enqueue(struct obd_export *exp,
                 RETURN(-EINVAL);
         }
 
-        mdc_pack_secdesc(req, reqsize[0]);
+        lustre_pack_secdesc(req, reqsize[0]);
 
         mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
         rc = ldlm_cli_enqueue(exp, req, obddev->obd_namespace, res_id,
@@ -321,13 +350,21 @@ int mdc_enqueue(struct obd_export *exp,
 
         /* This can go when we're sure that this can never happen */
         LASSERT(rc != -ENOENT);
+        /* We need dlm_rep to be assigned this early, to check lock mode of
+           returned lock from request to avoid possible race with lock
+           conversion */
+        if (rc == ELDLM_LOCK_ABORTED || !rc) {
+                dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
+                LASSERT(dlm_rep != NULL);   /* checked by ldlm_cli_enqueue() */
+        }
         if (rc == ELDLM_LOCK_ABORTED) {
                 lock_mode = 0;
                 memset(lockh, 0, sizeof(*lockh));
                 rc = 0;
         } else if (rc != 0) {
                 CERROR("ldlm_cli_enqueue: %d\n", rc);
-                LASSERT (rc < 0);
+                LASSERTF(rc < 0, "rc = %d\n", rc);
+                mdc_clear_replay_flag(req, rc);
                 ptlrpc_req_finished(req);
                 RETURN(rc);
         } else { /* rc = 0 */
@@ -336,43 +373,37 @@ int mdc_enqueue(struct obd_export *exp,
 
                 /* If the server gave us back a different lock mode, we should
                  * fix up our variables. */
-                if (lock->l_req_mode != lock_mode) {
-                        ldlm_lock_addref(lockh, lock->l_req_mode);
+                if (dlm_rep->lock_desc.l_req_mode != lock_mode) {
+                        ldlm_lock_addref(lockh, dlm_rep->lock_desc.l_req_mode);
                         ldlm_lock_decref(lockh, lock_mode);
-                        lock_mode = lock->l_req_mode;
+                        lock_mode = dlm_rep->lock_desc.l_req_mode;
                 }
 
                 ldlm_lock_allow_match(lock);
                 LDLM_LOCK_PUT(lock);
         }
 
-        dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
-        LASSERT(dlm_rep != NULL);           /* checked by ldlm_cli_enqueue() */
         LASSERT_REPSWABBED(req, 0);         /* swabbed by ldlm_cli_enqueue() */
 
-        it->d.lustre.it_disposition = (int) dlm_rep->lock_policy_res1;
-        it->d.lustre.it_status = (int) dlm_rep->lock_policy_res2;
-        it->d.lustre.it_lock_mode = lock_mode;
-        it->d.lustre.it_data = req;
+        LUSTRE_IT(it)->it_disposition = (int) dlm_rep->lock_policy_res1;
+        LUSTRE_IT(it)->it_status = (int) dlm_rep->lock_policy_res2;
+        LUSTRE_IT(it)->it_lock_mode = lock_mode;
+        LUSTRE_IT(it)->it_data = req;
 
-        if (it->d.lustre.it_status < 0 && req->rq_replay) {
-                LASSERT(req->rq_transno == 0);
-                /* Don't hold error requests for replay. */
-                spin_lock(&req->rq_lock);
-                req->rq_replay = 0;
-                spin_unlock(&req->rq_lock);
-        }
+        if (LUSTRE_IT(it)->it_status < 0 && req->rq_replay)
+                mdc_clear_replay_flag(req, LUSTRE_IT(it)->it_status);
 
-        DEBUG_REQ(D_RPCTRACE, req, "disposition: %x, status: %d",
-                  it->d.lustre.it_disposition, it->d.lustre.it_status);
+        DEBUG_REQ(D_DLMTRACE, req, "disposition: %x, status: %d",
+                  LUSTRE_IT(it)->it_disposition, LUSTRE_IT(it)->it_status);
 
         /* We know what to expect, so we do any byte flipping required here */
-        LASSERT(reply_buffers == 4 || reply_buffers == 3 || reply_buffers == 1);
+        LASSERT(reply_buffers == 5 || reply_buffers == 4 || 
+                reply_buffers == 3 || reply_buffers == 1);
         if (reply_buffers >= 3) {
                 struct mds_body *body;
 
                 body = lustre_swab_repbuf(req, 1, sizeof (*body),
-                                           lustre_swab_mds_body);
+                                          lustre_swab_mds_body);
                 if (body == NULL) {
                         CERROR ("Can't swab mds_body\n");
                         RETURN (-EPROTO);
@@ -424,73 +455,93 @@ EXPORT_SYMBOL(mdc_enqueue);
  * 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 d.lustre->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 d.lustre->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 d.lustre->it_status is the rc of the
  * child lookup.
  */
-int mdc_intent_lock(struct obd_export *exp,
-                    struct ll_fid *pfid, const char *name, int len,
-                    void *lmm, int lmmsize, struct ll_fid *cfid,
-                    struct lookup_intent *it, int lookup_flags,
-                    struct ptlrpc_request **reqp,
+int mdc_intent_lock(struct obd_export *exp, struct lustre_id *pid, 
+                    const char *name, int len, void *lmm, int lmmsize, 
+                    struct lustre_id *cid, struct lookup_intent *it, 
+                    int lookup_flags, struct ptlrpc_request **reqp,
                     ldlm_blocking_callback cb_blocking)
 {
         struct lustre_handle lockh;
         struct ptlrpc_request *request;
-        int rc = 0;
         struct mds_body *mds_body;
         struct lustre_handle old_lock;
         struct ldlm_lock *lock;
+        int rc = 0;
         ENTRY;
         LASSERT(it);
 
-        CDEBUG(D_DLMTRACE, "name: %*s in inode "LPU64", intent: %s flags %#o\n",
-               len, name, pfid->id, ldlm_it2str(it->it_op), it->it_flags);
+        CDEBUG(D_DLMTRACE, "name: %*s in obj "DLID4", intent: %s flags %#o\n",
+               len, name, OLID4(pid), ldlm_it2str(it->it_op), it->it_flags);
 
-        if (cfid && (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR ||
-                     it->it_op == IT_CHDIR)) {
+        if (cid && (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR ||
+                    it->it_op == IT_CHDIR)) {
                 /* 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 ={.name = {cfid->id,
-                                                     cfid->generation}};
+                struct ldlm_res_id res_id = {.name = {id_fid(cid),
+                                                      id_group(cid)}};
                 struct lustre_handle lockh;
                 ldlm_policy_data_t policy;
-                int mode = LCK_PR;
+                int mode;
 
                 /* For the GETATTR case, ll_revalidate_it issues two separate
-                   queries - for LOOKUP and for UPDATE lock because if cannot
+                   queries - for LOOKUP and for UPDATE lock because it cannot
                    check them together - we might have those two bits to be
                    present in two separate granted locks */
-                policy.l_inodebits.bits = 
-                                 (it->it_op == IT_GETATTR)?MDS_INODELOCK_UPDATE:
-                                                           MDS_INODELOCK_LOOKUP;
+                policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
+                        MDS_INODELOCK_UPDATE : MDS_INODELOCK_LOOKUP;
+                
                 mode = LCK_PR;
                 rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
                                      LDLM_FL_BLOCK_GRANTED, &res_id,
-                                     LDLM_IBITS, &policy, LCK_PR, &lockh);
+                                     LDLM_IBITS, &policy, mode,
+                                     &lockh);
+
+                if (!rc) {
+                        mode = LCK_CR;
+                        rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
+                                             LDLM_FL_BLOCK_GRANTED, &res_id,
+                                             LDLM_IBITS, &policy, mode,
+                                             &lockh);
+                }
                 if (!rc) {
                         mode = LCK_PW;
                         rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
                                              LDLM_FL_BLOCK_GRANTED, &res_id,
-                                             LDLM_IBITS, &policy, LCK_PW,
+                                             LDLM_IBITS, &policy, mode,
+                                             &lockh);
+                }
+                if (!rc) {
+                        mode = LCK_CW;
+                        rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
+                                             LDLM_FL_BLOCK_GRANTED, &res_id,
+                                             LDLM_IBITS, &policy, mode,
                                              &lockh);
                 }
                 if (rc) {
-                        memcpy(&it->d.lustre.it_lock_handle, &lockh,
+                        if (ptlrpcs_check_cred(exp->exp_obd->u.cli.cl_import)) {
+                                /* return immediately if no credential held */
+                                ldlm_lock_decref(&lockh, mode);
+                                RETURN(-EACCES);
+                        }
+                        memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh,
                                sizeof(lockh));
-                        it->d.lustre.it_lock_mode = mode;
+                        LUSTRE_IT(it)->it_lock_mode = mode;
                 }
 
-                /* Only return failure if it was not GETATTR by cfid
-                   (from inode_revalidate) */
+                /* Only return failure if it was not GETATTR by cid (from
+                   inode_revalidate) */
                 if (rc || name)
                         RETURN(rc);
         }
@@ -504,61 +555,63 @@ int mdc_intent_lock(struct obd_export *exp,
          * this and use the request from revalidate.  In this case, revalidate
          * never dropped its reference, so the refcounts are all OK */
         if (!it_disposition(it, DISP_ENQ_COMPLETE)) {
-                struct mdc_op_data op_data;
-                mdc_fid2mdc_op_data(&op_data, pfid, cfid, name, len, 0);
+                struct mdc_op_data *op_data;
+
+                OBD_ALLOC(op_data, sizeof(*op_data));
+                if (op_data == NULL)
+                        RETURN(-ENOMEM);
+
+                mdc_id2mdc_data(op_data, pid, cid, name, len, 0);
+
+                if (name != NULL)
+                        op_data->valid |= OBD_MD_FID;
 
                 rc = mdc_enqueue(exp, LDLM_IBITS, it, it_to_lock_mode(it),
-                                 &op_data, &lockh, lmm, lmmsize,
+                                 op_data, &lockh, lmm, lmmsize,
                                  ldlm_completion_ast, cb_blocking, NULL);
+                OBD_FREE(op_data, sizeof(*op_data));
                 if (rc < 0)
                         RETURN(rc);
-                memcpy(&it->d.lustre.it_lock_handle, &lockh, sizeof(lockh));
+                
+                memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh, sizeof(lockh));
         }
-        request = *reqp = it->d.lustre.it_data;
+        request = *reqp = LUSTRE_IT(it)->it_data;
         LASSERT(request != NULL);
         
         /* 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
-         * this request for unconditional replay.
+         * successful open, then we need to remove the bit which saves this
+         * request for unconditional replay.
          *
          * It's important that we do this first!  Otherwise we might exit the
-         * function without doing so, and try to replay a failed create
-         * (bug 3440) */
-        if (it->it_op & IT_OPEN) {
-                if (!it_disposition(it, DISP_OPEN_OPEN) ||
-                    it->d.lustre.it_status != 0) {
-                        unsigned long irqflags;
-
-                        spin_lock_irqsave(&request->rq_lock, irqflags);
-                        request->rq_replay = 0;
-                        spin_unlock_irqrestore(&request->rq_lock, irqflags);
-                }
-        }
+         * function without doing so, and try to replay a failed create (bug
+         * 3440) */
+        if (it->it_op & IT_OPEN && request->rq_replay &&
+            (!it_disposition(it, DISP_OPEN_OPEN) || LUSTRE_IT(it)->it_status != 0))
+                mdc_clear_replay_flag(request, LUSTRE_IT(it)->it_status);
         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);
+                LASSERT(LUSTRE_IT(it)->it_status != 0);
+                RETURN(LUSTRE_IT(it)->it_status);
         }
         rc = it_open_error(DISP_IT_EXECD, it);
         if (rc)
                 RETURN(rc);
 
         mds_body = lustre_msg_buf(request->rq_repmsg, 1, sizeof(*mds_body));
-        LASSERT(mds_body != NULL);           /* mdc_enqueue checked */
+        LASSERT(mds_body != NULL);      /* mdc_enqueue checked */
         LASSERT_REPSWABBED(request, 1); /* mdc_enqueue swabbed */
 
-        /* If we were revalidating a fid/name pair, mark the intent in
-         * case we fail and get called again from lookup */
-        if (cfid != NULL) {
+        /* If we were revalidating a fid/name pair, mark the intent in case we
+         * fail and get called again from lookup */
+        if (cid != NULL) {
                 it_set_disposition(it, DISP_ENQ_COMPLETE);
                 /* Also: did we find the same inode? */
-                /* we have to compare all the fields but type, because
-                 * MDS can return mds/ino/generation triple if inode
-                 * lives on another MDS -bzzz */
-                if (cfid->generation != mds_body->fid1.generation ||
-                                cfid->id != mds_body->fid1.id ||
-                                cfid->mds != mds_body->fid1.mds)
+                
+                /* we have to compare all the fields but type, because MDS can
+                 * return fid/mds/ino/gen if inode lives on another MDS -bzzz */
+                if (!(lookup_flags & LOOKUP_COBD) && !id_equal(cid, &mds_body->id1))
                         RETURN(-ESTALE);
         }
 
@@ -566,8 +619,9 @@ int mdc_intent_lock(struct obd_export *exp,
         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
+        /*
+         * 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_OPEN_CREATE) &&
             !it_open_error(DISP_OPEN_CREATE, it))
@@ -584,29 +638,40 @@ int mdc_intent_lock(struct obd_export *exp,
                 LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_CHDIR));
         }
 
-        /* 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().) */
+        /*
+         * 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");
                 LDLM_LOCK_PUT(lock);
+                
+                LASSERTF(id_fid(&mds_body->id1) == lock->l_resource->lr_name.name[0] &&
+                         id_group(&mds_body->id1) == lock->l_resource->lr_name.name[1],
+                         "Invalid lock is returned to client. Lock res_is: %lu/%lu, "
+                         "response res_id: %lu/%lu.\n",
+                         (unsigned long)lock->l_resource->lr_name.name[0],
+                         (unsigned long)lock->l_resource->lr_name.name[1],
+                         (unsigned long)id_fid(&mds_body->id1),
+                         (unsigned long)id_group(&mds_body->id1));
+                
                 memcpy(&old_lock, &lockh, sizeof(lockh));
                 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
                                     LDLM_IBITS, &policy, LCK_NL, &old_lock)) {
                         ldlm_lock_decref_and_cancel(&lockh,
-                                                    it->d.lustre.it_lock_mode);
+                                                    LUSTRE_IT(it)->it_lock_mode);
                         memcpy(&lockh, &old_lock, sizeof(old_lock));
-                        memcpy(&it->d.lustre.it_lock_handle, &lockh,
+                        memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh,
                                sizeof(lockh));
                 }
         }
         CDEBUG(D_DENTRY, "D_IT dentry %*s intent: %s status %d disp %x rc %d\n",
-               len, name, ldlm_it2str(it->it_op), it->d.lustre.it_status,
-               it->d.lustre.it_disposition, rc);
+               len, name, ldlm_it2str(it->it_op), LUSTRE_IT(it)->it_status,
+               LUSTRE_IT(it)->it_disposition, rc);
 
         RETURN(rc);
 }