Whamcloud - gitweb
LU-8371 llite: Trust creates in revalidate too.
[fs/lustre-release.git] / lustre / llite / dcache.c
index 88db91b..c42e89d 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 
 #include <linux/fs.h>
 #include <linux/sched.h>
-#include <linux/smp_lock.h>
 #include <linux/quotaops.h>
+#include <linux/kernel.h>
 
 #define DEBUG_SUBSYSTEM S_LLITE
 
 #include <obd_support.h>
-#include <lustre_lite.h>
 #include <lustre/lustre_idl.h>
 #include <lustre_dlm.h>
 
 #include "llite_internal.h"
 
+static void free_dentry_data(struct rcu_head *head)
+{
+       struct ll_dentry_data *lld;
+
+       lld = container_of(head, struct ll_dentry_data, lld_rcu_head);
+       OBD_FREE_PTR(lld);
+}
+
 /* should NOT be called with the dcache lock, see fs/dcache.c */
 static void ll_release(struct dentry *de)
 {
@@ -62,29 +69,40 @@ static void ll_release(struct dentry *de)
                 ll_intent_release(lld->lld_it);
                 OBD_FREE(lld->lld_it, sizeof(*lld->lld_it));
         }
-        LASSERT(lld->lld_cwd_count == 0);
-        LASSERT(lld->lld_mnt_count == 0);
-        OBD_FREE(de->d_fsdata, sizeof(*lld));
 
-        EXIT;
+       de->d_fsdata = NULL;
+       call_rcu(&lld->lld_rcu_head, free_dentry_data);
+
+       EXIT;
 }
 
 /* Compare if two dentries are the same.  Don't match if the existing dentry
- * is marked DCACHE_LUSTRE_INVALID.  Returns 1 if different, 0 if the same.
+ * is marked invalid.  Returns 1 if different, 0 if the same.
  *
  * This avoids a race where ll_lookup_it() instantiates a dentry, but we get
  * an AST before calling d_revalidate_it().  The dentry still exists (marked
  * INVALID) so d_lookup() matches it, but we have no lock on it (so
  * lock_match() fails) and we spin around real_lookup(). */
 #ifdef HAVE_D_COMPARE_7ARGS
-int ll_dcompare(const struct dentry *parent, const struct inode *pinode,
-               const struct dentry *dentry, const struct inode *inode,
-               unsigned int len, const char *str, const struct qstr *name)
+static int ll_dcompare(const struct dentry *parent, const struct inode *pinode,
+                      const struct dentry *dentry, const struct inode *inode,
+                      unsigned int len, const char *str,
+                      const struct qstr *name)
+#elif defined(HAVE_D_COMPARE_5ARGS)
+static int ll_dcompare(const struct dentry *parent, const struct dentry *dentry,
+                      unsigned int len, const char *str,
+                      const struct qstr *name)
 #else
-int ll_dcompare(struct dentry *parent, struct qstr *d_name, struct qstr *name)
+static int ll_dcompare(struct dentry *parent, struct qstr *d_name,
+                      struct qstr *name)
 #endif
 {
-#ifdef HAVE_D_COMPARE_7ARGS
+#if !defined(HAVE_D_COMPARE_7ARGS) && !defined(HAVE_D_COMPARE_5ARGS)
+       /* XXX: (ugh !) d_name must be in-dentry structure */
+       struct dentry *dentry = container_of(d_name, struct dentry, d_name);
+       unsigned int len = d_name->len;
+       const char *str = d_name->name;
+#endif
        ENTRY;
 
        if (len != name->len)
@@ -92,65 +110,19 @@ int ll_dcompare(struct dentry *parent, struct qstr *d_name, struct qstr *name)
 
        if (memcmp(str, name->name, len))
                RETURN(1);
-#else
-       struct dentry *dentry;
-       ENTRY;
-
-       if (d_name->len != name->len)
-               RETURN(1);
-
-       if (memcmp(d_name->name, name->name, name->len))
-               RETURN(1);
-
-       /* XXX: d_name must be in-dentry structure */
-       dentry = container_of(d_name, struct dentry, d_name); /* ugh */
-#endif
 
        CDEBUG(D_DENTRY, "found name %.*s(%p) flags %#x refc %d\n",
               name->len, name->name, dentry, dentry->d_flags,
-              d_refcount(dentry));
+              ll_d_count(dentry));
 
        /* mountpoint is always valid */
        if (d_mountpoint((struct dentry *)dentry))
                RETURN(0);
 
        if (d_lustre_invalid(dentry))
-                RETURN(1);
-
-        RETURN(0);
-}
-
-static inline int return_if_equal(struct ldlm_lock *lock, void *data)
-{
-        if ((lock->l_flags &
-             (LDLM_FL_CANCELING | LDLM_FL_DISCARD_DATA)) ==
-            (LDLM_FL_CANCELING | LDLM_FL_DISCARD_DATA))
-                return LDLM_ITER_CONTINUE;
-        return LDLM_ITER_STOP;
-}
-
-/* find any ldlm lock of the inode in mdc and lov
- * return 0    not find
- *        1    find one
- *      < 0    error */
-static int find_cbdata(struct inode *inode)
-{
-        struct ll_inode_info *lli = ll_i2info(inode);
-        struct ll_sb_info *sbi = ll_i2sbi(inode);
-        int rc = 0;
-        ENTRY;
-
-        LASSERT(inode);
-        rc = md_find_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
-                            return_if_equal, NULL);
-        if (rc != 0)
-                 RETURN(rc);
-
-        if (lli->lli_smd)
-                rc = obd_find_cbdata(sbi->ll_dt_exp, lli->lli_smd,
-                                     return_if_equal, NULL);
+               RETURN(1);
 
-        RETURN(rc);
+       RETURN(0);
 }
 
 /**
@@ -171,32 +143,26 @@ static int ll_ddelete(HAVE_D_DELETE_CONST struct dentry *de)
               d_unhashed((struct dentry *)de) ? "" : "hashed,",
               list_empty(&de->d_subdirs) ? "" : "subdirs");
 
-       /* kernel >= 2.6.38 last refcount is decreased after this function. */
-#ifdef DCACHE_OP_DELETE
-       LASSERT(d_refcount(de) == 1);
+#ifdef HAVE_DCACHE_LOCK
+       LASSERT(ll_d_count(de) == 0);
 #else
-       LASSERT(d_refcount(de) == 0);
+       /* kernel >= 2.6.38 last refcount is decreased after this function. */
+       LASSERT(ll_d_count(de) == 1);
 #endif
 
-       /* if not ldlm lock for this inode, set i_nlink to 0 so that
-        * this inode can be recycled later b=20433 */
-       if (de->d_inode && !find_cbdata(de->d_inode))
-               de->d_inode->i_nlink = 0;
-
        if (d_lustre_invalid((struct dentry *)de))
-                RETURN(1);
-
-        RETURN(0);
+               RETURN(1);
+       RETURN(0);
 }
 
-static int ll_set_dd(struct dentry *de)
+int ll_d_init(struct dentry *de)
 {
-        ENTRY;
-        LASSERT(de != NULL);
+       ENTRY;
+       LASSERT(de != NULL);
 
-        CDEBUG(D_DENTRY, "ldd on dentry %.*s (%p) parent %p inode %p refc %d\n",
-               de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode,
-               d_refcount(de));
+       CDEBUG(D_DENTRY, "ldd on dentry %.*s (%p) parent %p inode %p refc %d\n",
+               de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode,
+               ll_d_count(de));
 
        if (de->d_fsdata == NULL) {
                struct ll_dentry_data *lld;
@@ -204,57 +170,50 @@ static int ll_set_dd(struct dentry *de)
                OBD_ALLOC_PTR(lld);
                if (likely(lld != NULL)) {
                        spin_lock(&de->d_lock);
-                       if (likely(de->d_fsdata == NULL))
+                       if (likely(de->d_fsdata == NULL)) {
                                de->d_fsdata = lld;
-                       else
+                               __d_lustre_invalidate(de);
+#ifdef HAVE_DCACHE_LOCK
+                               /* kernel >= 2.6.38 d_op is set in d_alloc() */
+                               de->d_op = &ll_d_ops;
+#endif
+                       } else {
                                OBD_FREE_PTR(lld);
+                       }
                        spin_unlock(&de->d_lock);
-                } else {
-                        RETURN(-ENOMEM);
-                }
-        }
-
-        RETURN(0);
-}
-
-int ll_dops_init(struct dentry *de, int block, int init_sa)
-{
-        struct ll_dentry_data *lld = ll_d2d(de);
-        int rc = 0;
-
-        if (lld == NULL && block != 0) {
-                rc = ll_set_dd(de);
-                if (rc)
-                        return rc;
-
-                lld = ll_d2d(de);
-        }
-
-        if (lld != NULL && init_sa != 0)
-                lld->lld_sa_generation = 0;
-
-#ifdef DCACHE_OP_HASH
+               } else {
+                       RETURN(-ENOMEM);
+               }
+       }
        LASSERT(de->d_op == &ll_d_ops);
-#else
-       de->d_op = &ll_d_ops;
-#endif
-        return rc;
+
+       RETURN(0);
 }
 
 void ll_intent_drop_lock(struct lookup_intent *it)
 {
-        struct lustre_handle *handle;
-
-        if (it->it_op && it->d.lustre.it_lock_mode) {
-                handle = (struct lustre_handle *)&it->d.lustre.it_lock_handle;
-                CDEBUG(D_DLMTRACE, "releasing lock with cookie "LPX64
-                       " from it %p\n", handle->cookie, it);
-                ldlm_lock_decref(handle, it->d.lustre.it_lock_mode);
-
-                /* bug 494: intent_release may be called multiple times, from
-                 * this thread and we don't want to double-decref this lock */
-                it->d.lustre.it_lock_mode = 0;
-        }
+       if (it->it_op && it->it_lock_mode) {
+               struct lustre_handle handle;
+
+               handle.cookie = it->it_lock_handle;
+
+               CDEBUG(D_DLMTRACE, "releasing lock with cookie %#llx from it %p\n",
+                      handle.cookie, it);
+               ldlm_lock_decref(&handle, it->it_lock_mode);
+
+               /* bug 494: intent_release may be called multiple times, from
+                * this thread and we don't want to double-decref this lock */
+               it->it_lock_mode = 0;
+               if (it->it_remote_lock_mode != 0) {
+                       handle.cookie = it->it_remote_lock_handle;
+
+                       CDEBUG(D_DLMTRACE, "releasing remote lock with cookie"
+                              "%#llx from it %p\n", handle.cookie, it);
+                       ldlm_lock_decref(&handle,
+                                        it->it_remote_lock_mode);
+                       it->it_remote_lock_mode = 0;
+               }
+       }
 }
 
 void ll_intent_release(struct lookup_intent *it)
@@ -265,43 +224,43 @@ void ll_intent_release(struct lookup_intent *it)
         ll_intent_drop_lock(it);
         /* We are still holding extra reference on a request, need to free it */
         if (it_disposition(it, DISP_ENQ_OPEN_REF))
-                 ptlrpc_req_finished(it->d.lustre.it_data); /* ll_file_open */
-        if (it_disposition(it, DISP_ENQ_CREATE_REF)) /* create rec */
-                ptlrpc_req_finished(it->d.lustre.it_data);
-        if (it_disposition(it, DISP_ENQ_COMPLETE)) /* saved req from revalidate
-                                                    * to lookup */
-                ptlrpc_req_finished(it->d.lustre.it_data);
-
-        it->d.lustre.it_disposition = 0;
-        it->d.lustre.it_data = NULL;
-        EXIT;
+               ptlrpc_req_finished(it->it_request); /* ll_file_open */
+
+       if (it_disposition(it, DISP_ENQ_CREATE_REF)) /* create rec */
+               ptlrpc_req_finished(it->it_request);
+
+       it->it_disposition = 0;
+       it->it_request = NULL;
+       EXIT;
 }
 
 void ll_invalidate_aliases(struct inode *inode)
 {
        struct dentry *dentry;
+       DECLARE_LL_D_HLIST_NODE_PTR(p);
        ENTRY;
 
        LASSERT(inode != NULL);
 
-       CDEBUG(D_INODE, "marking dentries for ino %lu/%u(%p) invalid\n",
-              inode->i_ino, inode->i_generation, inode);
+       CDEBUG(D_INODE, "marking dentries for inode "DFID"(%p) invalid\n",
+              PFID(ll_inode2fid(inode)), inode);
 
        ll_lock_dcache(inode);
-       cfs_list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
-                CDEBUG(D_DENTRY, "dentry in drop %.*s (%p) parent %p "
-                       "inode %p flags %d\n", dentry->d_name.len,
-                       dentry->d_name.name, dentry, dentry->d_parent,
-                       dentry->d_inode, dentry->d_flags);
-
-                if (dentry->d_name.len == 1 && dentry->d_name.name[0] == '/') {
-                        CERROR("called on root (?) dentry=%p, inode=%p "
-                               "ino=%lu\n", dentry, inode, inode->i_ino);
-                        lustre_dump_dentry(dentry, 1);
+       ll_d_hlist_for_each_entry(dentry, p, &inode->i_dentry) {
+               CDEBUG(D_DENTRY, "dentry in drop %.*s (%p) parent %p "
+                      "inode %p flags %d\n", dentry->d_name.len,
+                      dentry->d_name.name, dentry, dentry->d_parent,
+                      dentry->d_inode, dentry->d_flags);
+
+               if (unlikely(dentry == dentry->d_sb->s_root)) {
+                       CERROR("%s: called on root dentry=%p, fid="DFID"\n",
+                              ll_get_fsname(dentry->d_sb, NULL, 0),
+                              dentry, PFID(ll_inode2fid(inode)));
+                       lustre_dump_dentry(dentry, 1);
                         libcfs_debug_dumpstack(NULL);
                 }
 
-               d_lustre_invalidate(dentry);
+               d_lustre_invalidate(dentry, 0);
        }
        ll_unlock_dcache(inode);
 
@@ -321,7 +280,7 @@ int ll_revalidate_it_finish(struct ptlrpc_request *request,
         if (it_disposition(it, DISP_LOOKUP_NEG))
                 RETURN(-ENOENT);
 
-        rc = ll_prep_inode(&de->d_inode, request, NULL);
+        rc = ll_prep_inode(&de->d_inode, request, NULL, it);
 
         RETURN(rc);
 }
@@ -331,12 +290,12 @@ void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry)
         LASSERT(it != NULL);
         LASSERT(dentry != NULL);
 
-        if (it->d.lustre.it_lock_mode && dentry->d_inode != NULL) {
+       if (it->it_lock_mode && dentry->d_inode != NULL) {
                 struct inode *inode = dentry->d_inode;
                 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
 
-                CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
-                       inode, inode->i_ino, inode->i_generation);
+               CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)\n",
+                      PFID(ll_inode2fid(inode)), inode);
                 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
         }
 
@@ -349,344 +308,85 @@ void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry)
         }
 }
 
-void ll_frob_intent(struct lookup_intent **itp, struct lookup_intent *deft)
+static int ll_revalidate_dentry(struct dentry *dentry,
+                               unsigned int lookup_flags)
 {
-        struct lookup_intent *it = *itp;
+       struct inode *dir = dentry->d_parent->d_inode;
+
+       /* If this is intermediate component path lookup and we were able to get
+        * to this dentry, then its lock has not been revoked and the
+        * path component is valid. */
+       if (lookup_flags & (LOOKUP_CONTINUE | LOOKUP_PARENT))
+               return 1;
 
-        if (!it || it->it_op == IT_GETXATTR)
-                it = *itp = deft;
+       /* Symlink - always valid as long as the dentry was found */
+#ifdef HAVE_IOP_GET_LINK
+       if (dentry->d_inode && dentry->d_inode->i_op->get_link)
+#else
+       if (dentry->d_inode && dentry->d_inode->i_op->follow_link)
+#endif
+               return 1;
+
+       /*
+        * VFS warns us that this is the second go around and previous
+        * operation failed (most likely open|creat), so this time
+        * we better talk to the server via the lookup path by name,
+        * not by fid.
+        */
+       if (lookup_flags & LOOKUP_REVAL)
+               return 0;
+
+       if (!dentry_may_statahead(dir, dentry))
+               return 1;
+
+#ifndef HAVE_DCACHE_LOCK
+       if (lookup_flags & LOOKUP_RCU)
+               return -ECHILD;
+#endif
 
+       ll_statahead(dir, &dentry, dentry->d_inode == NULL);
+       return 1;
 }
 
-int ll_revalidate_it(struct dentry *de, int lookup_flags,
-                     struct lookup_intent *it)
+/*
+ * Always trust cached dentries. Update statahead window if necessary.
+ */
+#ifdef HAVE_IOP_ATOMIC_OPEN
+static int ll_revalidate_nd(struct dentry *dentry, unsigned int flags)
 {
-        struct md_op_data *op_data;
-        struct ptlrpc_request *req = NULL;
-        struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
-        struct obd_export *exp;
-        struct inode *parent = de->d_parent->d_inode;
-        int rc;
-
-        ENTRY;
-        CDEBUG(D_VFSTRACE, "VFS Op:name=%s,intent=%s\n", de->d_name.name,
-               LL_IT2STR(it));
-
-        if (de->d_inode == NULL) {
-                __u64 ibits;
-
-                /* We can only use negative dentries if this is stat or lookup,
-                   for opens and stuff we do need to query server. */
-                /* If there is IT_CREAT in intent op set, then we must throw
-                   away this negative dentry and actually do the request to
-                   kernel to create whatever needs to be created (if possible)*/
-                if (it && (it->it_op & IT_CREAT))
-                        RETURN(0);
-
-               if (d_lustre_invalid(de))
-                        RETURN(0);
-
-                ibits = MDS_INODELOCK_UPDATE;
-                rc = ll_have_md_lock(parent, &ibits, LCK_MINMODE);
-                GOTO(out_sa, rc);
-        }
-
-        /* Never execute intents for mount points.
-         * Attributes will be fixed up in ll_inode_revalidate_it */
-        if (d_mountpoint(de))
-                GOTO(out_sa, rc = 1);
-
-        /* need to get attributes in case root got changed from other client */
-        if (de == de->d_sb->s_root) {
-                rc = __ll_inode_revalidate_it(de, it, MDS_INODELOCK_LOOKUP);
-                if (rc == 0)
-                        rc = 1;
-                GOTO(out_sa, rc);
-        }
-
-        exp = ll_i2mdexp(de->d_inode);
-
-        OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_REVALIDATE_PAUSE, 5);
-        ll_frob_intent(&it, &lookup_it);
-        LASSERT(it);
-
-       if (it->it_op == IT_LOOKUP && !d_lustre_invalid(de))
-                RETURN(1);
-
-        if ((it->it_op == IT_OPEN) && de->d_inode) {
-                struct inode *inode = de->d_inode;
-                struct ll_inode_info *lli = ll_i2info(inode);
-                struct obd_client_handle **och_p;
-                __u64 *och_usecount;
-                __u64 ibits;
-
-                /*
-                 * We used to check for MDS_INODELOCK_OPEN here, but in fact
-                 * just having LOOKUP lock is enough to justify inode is the
-                 * same. And if inode is the same and we have suitable
-                 * openhandle, then there is no point in doing another OPEN RPC
-                 * just to throw away newly received openhandle.  There are no
-                 * security implications too, if file owner or access mode is
-                 * change, LOOKUP lock is revoked.
-                 */
-
-
-                if (it->it_flags & FMODE_WRITE) {
-                        och_p = &lli->lli_mds_write_och;
-                        och_usecount = &lli->lli_open_fd_write_count;
-                } else if (it->it_flags & FMODE_EXEC) {
-                        och_p = &lli->lli_mds_exec_och;
-                        och_usecount = &lli->lli_open_fd_exec_count;
-                } else {
-                        och_p = &lli->lli_mds_read_och;
-                        och_usecount = &lli->lli_open_fd_read_count;
-                }
-                /* Check for the proper lock. */
-                ibits = MDS_INODELOCK_LOOKUP;
-                if (!ll_have_md_lock(inode, &ibits, LCK_MINMODE))
-                        goto do_lock;
-                cfs_mutex_lock(&lli->lli_och_mutex);
-                if (*och_p) { /* Everything is open already, do nothing */
-                        /*(*och_usecount)++;  Do not let them steal our open
-                          handle from under us */
-                        SET_BUT_UNUSED(och_usecount);
-                        /* XXX The code above was my original idea, but in case
-                           we have the handle, but we cannot use it due to later
-                           checks (e.g. O_CREAT|O_EXCL flags set), nobody
-                           would decrement counter increased here. So we just
-                           hope the lock won't be invalidated in between. But
-                           if it would be, we'll reopen the open request to
-                           MDS later during file open path */
-                        cfs_mutex_unlock(&lli->lli_och_mutex);
-                        RETURN(1);
-                } else {
-                        cfs_mutex_unlock(&lli->lli_och_mutex);
-                }
-        }
-
-        if (it->it_op == IT_GETATTR) {
-                rc = ll_statahead_enter(parent, &de, 0);
-                if (rc == 1)
-                        goto mark;
-                else if (rc != -EAGAIN && rc != 0)
-                        GOTO(out, rc = 0);
-        }
-
-do_lock:
-        op_data = ll_prep_md_op_data(NULL, parent, de->d_inode,
-                                     de->d_name.name, de->d_name.len,
-                                     0, LUSTRE_OPC_ANY, NULL);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
-
-        it->it_create_mode &= ~cfs_curproc_umask();
-        it->it_create_mode |= M_CHECK_STALE;
-        rc = md_intent_lock(exp, op_data, NULL, 0, it,
-                            lookup_flags,
-                            &req, ll_md_blocking_ast, 0);
-        it->it_create_mode &= ~M_CHECK_STALE;
-        ll_finish_md_op_data(op_data);
-
-        /* If req is NULL, then md_intent_lock only tried to do a lock match;
-         * if all was well, it will return 1 if it found locks, 0 otherwise. */
-        if (req == NULL && rc >= 0) {
-                if (!rc)
-                        goto do_lookup;
-                GOTO(out, rc);
-        }
-
-        if (rc < 0) {
-                if (rc != -ESTALE) {
-                        CDEBUG(D_INFO, "ll_intent_lock: rc %d : it->it_status "
-                               "%d\n", rc, it->d.lustre.it_status);
-                }
-                GOTO(out, rc = 0);
-        }
-
-revalidate_finish:
-        rc = ll_revalidate_it_finish(req, it, de);
-        if (rc != 0) {
-                if (rc != -ESTALE && rc != -ENOENT)
-                        ll_intent_release(it);
-                GOTO(out, rc = 0);
-        }
-
-        if ((it->it_op & IT_OPEN) && de->d_inode &&
-            !S_ISREG(de->d_inode->i_mode) &&
-            !S_ISDIR(de->d_inode->i_mode)) {
-                ll_release_openhandle(de, it);
-        }
-        rc = 1;
-
-out:
-        /* We do not free request as it may be reused during following lookup
-         * (see comment in mdc/mdc_locks.c::mdc_intent_lock()), request will
-         * be freed in ll_lookup_it or in ll_intent_release. But if
-         * request was not completed, we need to free it. (bug 5154, 9903) */
-        if (req != NULL && !it_disposition(it, DISP_ENQ_COMPLETE))
-                ptlrpc_req_finished(req);
-        if (rc == 0) {
-               ll_invalidate_aliases(de->d_inode);
-       } else {
-               __u64 bits = 0;
-
-               CDEBUG(D_DENTRY, "revalidated dentry %.*s (%p) parent %p "
-                      "inode %p refc %d\n", de->d_name.len,
-                      de->d_name.name, de, de->d_parent, de->d_inode,
-                      d_refcount(de));
-               ll_set_lock_data(exp, de->d_inode, it, &bits);
-               if ((bits & MDS_INODELOCK_LOOKUP) && d_lustre_invalid(de))
-                       d_lustre_revalidate(de);
-                ll_lookup_finish_locks(it, de);
-        }
-
-mark:
-        if (it != NULL && it->it_op == IT_GETATTR && rc > 0)
-                ll_statahead_mark(parent, de);
-        RETURN(rc);
+       int rc;
+       ENTRY;
 
-        /*
-         * This part is here to combat evil-evil race in real_lookup on 2.6
-         * kernels.  The race details are: We enter do_lookup() looking for some
-         * name, there is nothing in dcache for this name yet and d_lookup()
-         * returns NULL.  We proceed to real_lookup(), and while we do this,
-         * another process does open on the same file we looking up (most simple
-         * reproducer), open succeeds and the dentry is added. Now back to
-         * us. In real_lookup() we do d_lookup() again and suddenly find the
-         * dentry, so we call d_revalidate on it, but there is no lock, so
-         * without this code we would return 0, but unpatched real_lookup just
-         * returns -ENOENT in such a case instead of retrying the lookup. Once
-         * this is dealt with in real_lookup(), all of this ugly mess can go and
-         * we can just check locks in ->d_revalidate without doing any RPCs
-         * ever.
-         */
-do_lookup:
-        if (it != &lookup_it) {
-                /* MDS_INODELOCK_UPDATE needed for IT_GETATTR case. */
-                if (it->it_op == IT_GETATTR)
-                        lookup_it.it_op = IT_GETATTR;
-                ll_lookup_finish_locks(it, de);
-                it = &lookup_it;
-        }
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%s, flags=%u\n",
+              dentry->d_name.name, flags);
 
-        /* Do real lookup here. */
-        op_data = ll_prep_md_op_data(NULL, parent, NULL, de->d_name.name,
-                                     de->d_name.len, 0, (it->it_op & IT_CREAT ?
-                                                         LUSTRE_OPC_CREATE :
-                                                         LUSTRE_OPC_ANY), NULL);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
-
-        rc = md_intent_lock(exp, op_data, NULL, 0,  it, 0, &req,
-                            ll_md_blocking_ast, 0);
-        if (rc >= 0) {
-                struct mdt_body *mdt_body;
-                struct lu_fid fid = {.f_seq = 0, .f_oid = 0, .f_ver = 0};
-                mdt_body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-
-                if (de->d_inode)
-                        fid = *ll_inode2fid(de->d_inode);
-
-                /* see if we got same inode, if not - return error */
-                if (lu_fid_eq(&fid, &mdt_body->fid1)) {
-                        ll_finish_md_op_data(op_data);
-                        op_data = NULL;
-                        goto revalidate_finish;
-                }
-                ll_intent_release(it);
-        }
-        ll_finish_md_op_data(op_data);
-        GOTO(out, rc = 0);
-
-out_sa:
-        /*
-         * For rc == 1 case, should not return directly to prevent losing
-         * statahead windows; for rc == 0 case, the "lookup" will be done later.
-         */
-        if (it != NULL && it->it_op == IT_GETATTR && rc == 1)
-                ll_statahead_enter(parent, &de, 1);
-        goto mark;
+       rc = ll_revalidate_dentry(dentry, flags);
+       RETURN(rc);
 }
-
-int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd)
+#else
+static int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd)
 {
-        int rc;
-        ENTRY;
-
-#ifdef LOOKUP_RCU
-       if (nd->flags & LOOKUP_RCU)
-               return -ECHILD;
-#endif
-
-        if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
-                struct lookup_intent *it;
-
-                it = ll_convert_intent(&nd->intent.open, nd->flags);
-                if (IS_ERR(it))
-                        RETURN(0);
-
-                if (it->it_op == (IT_OPEN|IT_CREAT) &&
-                    nd->intent.open.flags & O_EXCL) {
-                        CDEBUG(D_VFSTRACE, "create O_EXCL, returning 0\n");
-                        rc = 0;
-                        goto out_it;
-                }
-
-                rc = ll_revalidate_it(dentry, nd->flags, it);
-
-                if (rc && (nd->flags & LOOKUP_OPEN) &&
-                    it_disposition(it, DISP_OPEN_OPEN)) {/*Open*/
-// XXX Code duplication with ll_lookup_nd
-                        if (S_ISFIFO(dentry->d_inode->i_mode)) {
-                                // We cannot call open here as it would
-                                // deadlock.
-                                ptlrpc_req_finished(
-                                               (struct ptlrpc_request *)
-                                                  it->d.lustre.it_data);
-                        } else {
-                               struct file *filp;
-
-                               nd->intent.open.file->private_data = it;
-                               filp = lookup_instantiate_filp(nd, dentry,NULL);
-                               if (IS_ERR(filp))
-                                       rc = PTR_ERR(filp);
-                        }
-                }
-                if (!rc && (nd->flags & LOOKUP_CREATE) &&
-                    it_disposition(it, DISP_OPEN_CREATE)) {
-                        /* We created something but we may only return
-                         * negative dentry here, so save request in dentry,
-                         * if lookup will be called later on, it will
-                         * pick the request, otherwise it would be freed
-                         * with dentry */
-                        ll_d2d(dentry)->lld_it = it;
-                        it = NULL; /* avoid freeing */
-                }
+       int rc;
+       ENTRY;
 
-out_it:
-                if (it) {
-                        ll_intent_release(it);
-                        OBD_FREE(it, sizeof(*it));
-                }
-        } else {
-                rc = ll_revalidate_it(dentry, 0, NULL);
-        }
+       /*
+        * this is normally called from NFS export, and we don't know whether
+        * this is the last component.
+        */
+       if (nd == NULL)
+               RETURN(1);
 
-        RETURN(rc);
-}
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%s, flags=%u\n",
+              dentry->d_name.name, nd->flags);
 
-void ll_d_iput(struct dentry *de, struct inode *inode)
-{
-        LASSERT(inode);
-        if (!find_cbdata(inode))
-                inode->i_nlink = 0;
-        iput(inode);
+       rc = ll_revalidate_dentry(dentry, nd->flags);
+       RETURN(rc);
 }
+#endif
 
-struct dentry_operations ll_d_ops = {
+const struct dentry_operations ll_d_ops = {
         .d_revalidate = ll_revalidate_nd,
         .d_release = ll_release,
         .d_delete  = ll_ddelete,
-        .d_iput    = ll_d_iput,
         .d_compare = ll_dcompare,
 };