Whamcloud - gitweb
LU-4429 llite: fix open lock matching in ll_md_blocking_ast()
[fs/lustre-release.git] / lustre / llite / namei.c
index ee51be1..b3780e2 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -28,6 +26,8 @@
 /*
  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2013, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #include <linux/fs.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
-#include <linux/smp_lock.h>
 #include <linux/quotaops.h>
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
+#include <linux/security.h>
 
 #define DEBUG_SUBSYSTEM S_LLITE
 
 #include <lustre_lite.h>
 #include <lustre_dlm.h>
 #include <lustre_ver.h>
-#include <lustre_mdc.h>
 #include "llite_internal.h"
 
+static int ll_create_it(struct inode *, struct dentry *,
+                       int, struct lookup_intent *);
+
 /*
  * Check if we have something mounted at the named dchild.
  * In such a case there would always be dentry present.
@@ -111,6 +113,20 @@ static int ll_set_inode(struct inode *inode, void *opaque)
         }
 
         lli->lli_fid = body->fid1;
+        if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
+                CERROR("Can not initialize inode "DFID" without object type: "
+                       "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
+                return -EINVAL;
+        }
+
+        inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
+        if (unlikely(inode->i_mode == 0)) {
+                CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
+                return -EINVAL;
+        }
+
+        ll_lli_init(lli);
+
         return 0;
 }
 
@@ -130,12 +146,18 @@ struct inode *ll_iget(struct super_block *sb, ino_t hash,
 
         if (inode) {
                 if (inode->i_state & I_NEW) {
-                        int rc;
+                        int rc = 0;
 
                         ll_read_inode2(inode, md);
-                        rc = cl_inode_init(inode, md);
+                        if (S_ISREG(inode->i_mode) &&
+                            ll_i2info(inode)->lli_clob == NULL) {
+                               CDEBUG(D_INODE,
+                                       "%s: apply lsm %p to inode "DFID".\n",
+                                       ll_get_fsname(sb, NULL, 0), md->lsm,
+                                       PFID(ll_inode2fid(inode)));
+                                rc = cl_file_inode_init(inode, md);
+                       }
                         if (rc != 0) {
-                                md->lsm = NULL;
                                 make_bad_inode(inode);
                                 unlock_new_inode(inode);
                                 iput(inode);
@@ -143,151 +165,159 @@ struct inode *ll_iget(struct super_block *sb, ino_t hash,
                         } else
                                 unlock_new_inode(inode);
                 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
-                                ll_update_inode(inode, md);
-                CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
-                       inode, PFID(&md->body->fid1));
+                        ll_update_inode(inode, md);
+                       CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p)\n",
+                              PFID(&md->body->fid1), inode);
         }
         RETURN(inode);
 }
 
-static void ll_drop_negative_dentry(struct inode *dir)
+static void ll_invalidate_negative_children(struct inode *dir)
 {
-        struct dentry *dentry, *tmp_alias, *tmp_subdir;
-
-        cfs_spin_lock(&ll_lookup_lock);
-        spin_lock(&dcache_lock);
-restart:
-        list_for_each_entry_safe(dentry, tmp_alias,
-                                 &dir->i_dentry,d_alias) {
-                if (!list_empty(&dentry->d_subdirs)) {
-                        struct dentry *child;
-                        list_for_each_entry_safe(child, tmp_subdir,
-                                                 &dentry->d_subdirs,
-                                                 d_child) {
-                                /* XXX Print some debug here? */
-                                if (!child->d_inode)
-                                /* Negative dentry. If we were
-                                   dropping dcache lock, go
-                                   throught the list again */
-                                        if (ll_drop_dentry(child))
-                                                goto restart;
-                        }
-                }
-        }
-        spin_unlock(&dcache_lock);
-        cfs_spin_unlock(&ll_lookup_lock);
+       struct dentry *dentry, *tmp_subdir;
+       DECLARE_LL_D_HLIST_NODE_PTR(p);
+
+       ll_lock_dcache(dir);
+       ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry, d_alias) {
+               spin_lock(&dentry->d_lock);
+               if (!list_empty(&dentry->d_subdirs)) {
+                       struct dentry *child;
+
+                       list_for_each_entry_safe(child, tmp_subdir,
+                                                &dentry->d_subdirs,
+                                                d_u.d_child) {
+                               if (child->d_inode == NULL)
+                                       d_lustre_invalidate(child, 1);
+                       }
+               }
+               spin_unlock(&dentry->d_lock);
+       }
+       ll_unlock_dcache(dir);
 }
 
-
 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
-                       void *data, int flag)
+                      void *data, int flag)
 {
-        int rc;
-        struct lustre_handle lockh;
-        ENTRY;
-
-        switch (flag) {
-        case LDLM_CB_BLOCKING:
-                ldlm_lock2handle(lock, &lockh);
-                rc = ldlm_cli_cancel(&lockh);
-                if (rc < 0) {
-                        CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
-                        RETURN(rc);
-                }
-                break;
-        case LDLM_CB_CANCELING: {
-                struct inode *inode = ll_inode_from_lock(lock);
-                struct ll_inode_info *lli;
-                __u64 bits = lock->l_policy_data.l_inodebits.bits;
-                struct lu_fid *fid;
-
-                /* Invalidate all dentries associated with this inode */
-                if (inode == NULL)
-                        break;
-
-                LASSERT(lock->l_flags & LDLM_FL_CANCELING);
-                if ((bits & MDS_INODELOCK_LOOKUP) &&
-                    ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
-                        bits &= ~MDS_INODELOCK_LOOKUP;
-                if ((bits & MDS_INODELOCK_UPDATE) &&
-                    ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
-                        bits &= ~MDS_INODELOCK_UPDATE;
-                if ((bits & MDS_INODELOCK_OPEN) &&
-                    ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
-                        bits &= ~MDS_INODELOCK_OPEN;
-
-                fid = ll_inode2fid(inode);
-                if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
-                    lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
-                    lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
-                        LDLM_ERROR(lock, "data mismatch with object "
-                                   DFID" (%p)", PFID(fid), inode);
-                }
-
-                if (bits & MDS_INODELOCK_OPEN) {
-                        int flags = 0;
-                        switch (lock->l_req_mode) {
-                        case LCK_CW:
-                                flags = FMODE_WRITE;
-                                break;
-                        case LCK_PR:
-                                flags = FMODE_EXEC;
-                                break;
-                        case LCK_CR:
-                                flags = FMODE_READ;
-                                break;
-                        default:
-                                CERROR("Unexpected lock mode for OPEN lock "
-                                       "%d, inode %ld\n", lock->l_req_mode,
-                                       inode->i_ino);
-                        }
-                        ll_md_real_close(inode, flags);
-                }
-
-                lli = ll_i2info(inode);
-                if (bits & MDS_INODELOCK_UPDATE)
-                        lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
-
-                if (S_ISDIR(inode->i_mode) &&
-                     (bits & MDS_INODELOCK_UPDATE)) {
-                        CDEBUG(D_INODE, "invalidating inode %lu\n",
-                               inode->i_ino);
-                        truncate_inode_pages(inode->i_mapping, 0);
-                        ll_drop_negative_dentry(inode);
-                }
-
-                if ((bits & MDS_INODELOCK_LOOKUP) &&
-                    !cfs_list_empty(&lli->lli_sa_dentry)) {
-                        struct ll_dentry_data *lld, *next;
-
-                        cfs_spin_lock(&lli->lli_sa_lock);
-                        cfs_list_for_each_entry_safe(lld, next,
-                                                     &lli->lli_sa_dentry,
-                                                     lld_sa_alias)
-                                cfs_list_del_init(&lld->lld_sa_alias);
-                        cfs_spin_unlock(&lli->lli_sa_lock);
-                }
-
-                if (inode->i_sb->s_root &&
-                    inode != inode->i_sb->s_root->d_inode &&
-                    (bits & MDS_INODELOCK_LOOKUP))
-                        ll_unhash_aliases(inode);
-                iput(inode);
-                break;
-        }
-        default:
-                LBUG();
-        }
-
-        RETURN(0);
+       struct lustre_handle lockh;
+       int rc;
+       ENTRY;
+
+       switch (flag) {
+       case LDLM_CB_BLOCKING:
+               ldlm_lock2handle(lock, &lockh);
+               rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
+               if (rc < 0) {
+                       CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
+                       RETURN(rc);
+               }
+               break;
+       case LDLM_CB_CANCELING: {
+               struct inode *inode = ll_inode_from_resource_lock(lock);
+               __u64 bits = lock->l_policy_data.l_inodebits.bits;
+
+               /* Inode is set to lock->l_resource->lr_lvb_inode
+                * for mdc - bug 24555 */
+               LASSERT(lock->l_ast_data == NULL);
+
+               if (inode == NULL)
+                       break;
+
+               /* Invalidate all dentries associated with this inode */
+               LASSERT(ldlm_is_canceling(lock));
+
+               if (!fid_res_name_eq(ll_inode2fid(inode),
+                                    &lock->l_resource->lr_name)) {
+                       LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
+                                  PFID(ll_inode2fid(inode)), inode);
+                       LBUG();
+               }
+
+               if (bits & MDS_INODELOCK_XATTR) {
+                       ll_xattr_cache_destroy(inode);
+                       bits &= ~MDS_INODELOCK_XATTR;
+               }
+
+               /* For OPEN locks we differentiate between lock modes
+                * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
+               if (bits & MDS_INODELOCK_OPEN)
+                       ll_have_md_lock(inode, &bits, lock->l_req_mode);
+
+               if (bits & MDS_INODELOCK_OPEN) {
+                       fmode_t fmode;
+
+                       switch (lock->l_req_mode) {
+                       case LCK_CW:
+                               fmode = FMODE_WRITE;
+                               break;
+                       case LCK_PR:
+                               fmode = FMODE_EXEC;
+                               break;
+                       case LCK_CR:
+                               fmode = FMODE_READ;
+                               break;
+                       default:
+                               LDLM_ERROR(lock, "bad lock mode for OPEN lock");
+                               LBUG();
+                       }
+
+                       ll_md_real_close(inode, fmode);
+
+                       bits &= ~MDS_INODELOCK_OPEN;
+               }
+
+               if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
+                           MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM))
+                       ll_have_md_lock(inode, &bits, LCK_MINMODE);
+
+               if (bits & MDS_INODELOCK_LAYOUT) {
+                       struct cl_object_conf conf = {
+                               .coc_opc = OBJECT_CONF_INVALIDATE,
+                               .coc_inode = inode,
+                       };
+
+                       rc = ll_layout_conf(inode, &conf);
+                       if (rc < 0)
+                               CDEBUG(D_INODE, "cannot invalidate layout of "
+                                      DFID": rc = %d\n",
+                                      PFID(ll_inode2fid(inode)), rc);
+               }
+
+               if (bits & MDS_INODELOCK_UPDATE) {
+                       struct ll_inode_info *lli = ll_i2info(inode);
+
+                       spin_lock(&lli->lli_lock);
+                       lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
+                       spin_unlock(&lli->lli_lock);
+               }
+
+               if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
+                       CDEBUG(D_INODE, "invalidating inode "DFID"\n",
+                              PFID(ll_inode2fid(inode)));
+                       truncate_inode_pages(inode->i_mapping, 0);
+                       ll_invalidate_negative_children(inode);
+               }
+
+               if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
+                   inode->i_sb->s_root != NULL &&
+                   inode != inode->i_sb->s_root->d_inode)
+                       ll_invalidate_aliases(inode);
+
+               iput(inode);
+               break;
+       }
+       default:
+               LBUG();
+       }
+
+       RETURN(0);
 }
 
 __u32 ll_i2suppgid(struct inode *i)
 {
-        if (cfs_curproc_is_in_groups(i->i_gid))
-                return (__u32)i->i_gid;
-        else
-                return (__u32)(-1);
+       if (in_group_p(i->i_gid))
+               return (__u32)i->i_gid;
+       else
+               return (__u32)(-1);
 }
 
 /* Pack the required supplementary groups into the supplied groups array.
@@ -327,169 +357,109 @@ void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 #endif
 }
 
-static void ll_d_add(struct dentry *de, struct inode *inode)
+/*
+ * try to reuse three types of dentry:
+ * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
+ *    by concurrent .revalidate).
+ * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
+ *    be cleared by others calling d_lustre_revalidate).
+ * 3. DISCONNECTED alias.
+ */
+static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
 {
-        CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
-        /* d_instantiate */
-        if (!list_empty(&de->d_alias)) {
-                spin_unlock(&dcache_lock);
-                CERROR("dentry %.*s %p alias next %p, prev %p\n",
-                       de->d_name.len, de->d_name.name, de,
-                       de->d_alias.next, de->d_alias.prev);
-                LBUG();
-        }
-        if (inode)
-                list_add(&de->d_alias, &inode->i_dentry);
-        de->d_inode = inode;
-
-        /* d_rehash */
-        if (!d_unhashed(de)) {
-                spin_unlock(&dcache_lock);
-                CERROR("dentry %.*s %p hash next %p\n",
-                       de->d_name.len, de->d_name.name, de, de->d_hash.next);
-                LBUG();
-        }
-        d_rehash_cond(de, 0);
+       struct dentry *alias, *discon_alias, *invalid_alias;
+       DECLARE_LL_D_HLIST_NODE_PTR(p);
+
+       if (ll_d_hlist_empty(&inode->i_dentry))
+               return NULL;
+
+       discon_alias = invalid_alias = NULL;
+
+       ll_lock_dcache(inode);
+       ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
+               LASSERT(alias != dentry);
+
+               spin_lock(&alias->d_lock);
+               if (alias->d_flags & DCACHE_DISCONNECTED)
+                       /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
+                       discon_alias = alias;
+               else if (alias->d_parent == dentry->d_parent             &&
+                        alias->d_name.hash == dentry->d_name.hash       &&
+                        alias->d_name.len == dentry->d_name.len         &&
+                        memcmp(alias->d_name.name, dentry->d_name.name,
+                               dentry->d_name.len) == 0)
+                       invalid_alias = alias;
+               spin_unlock(&alias->d_lock);
+
+               if (invalid_alias)
+                       break;
+       }
+       alias = invalid_alias ?: discon_alias ?: NULL;
+       if (alias) {
+               spin_lock(&alias->d_lock);
+               dget_dlock(alias);
+               spin_unlock(&alias->d_lock);
+       }
+       ll_unlock_dcache(inode);
+
+       return alias;
 }
 
-/* Search "inode"'s alias list for a dentry that has the same name and parent
- * as de.  If found, return it.  If not found, return de.
- * Lustre can't use d_add_unique because don't unhash aliases for directory
- * in ll_revalidate_it.  After revaliadate inode will be have hashed aliases
- * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
+/*
+ * Similar to d_splice_alias(), but lustre treats invalid alias
+ * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
  */
-static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
+struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
 {
-        struct list_head *tmp;
-        struct dentry *dentry;
-        struct dentry *last_discon = NULL;
-
-        cfs_spin_lock(&ll_lookup_lock);
-        spin_lock(&dcache_lock);
-        list_for_each(tmp, &inode->i_dentry) {
-                dentry = list_entry(tmp, struct dentry, d_alias);
-
-                /* We are called here with 'de' already on the aliases list. */
-                if (unlikely(dentry == de)) {
-                        CERROR("whoops\n");
-                        continue;
-                }
-
-                if (dentry->d_flags & DCACHE_DISCONNECTED) {
-                        LASSERT(last_discon == NULL);
-                        last_discon = dentry;
-                        continue;
-                }
-
-                if (dentry->d_parent != de->d_parent)
-                        continue;
-
-                if (dentry->d_name.hash != de->d_name.hash)
-                        continue;
-
-                if (dentry->d_name.len != de->d_name.len)
-                        continue;
-
-                if (memcmp(dentry->d_name.name, de->d_name.name,
-                           de->d_name.len) != 0)
-                        continue;
-
-                dget_locked(dentry);
-                lock_dentry(dentry);
-                __d_drop(dentry);
-                unlock_dentry(dentry);
-                ll_dops_init(dentry, 0);
-                d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
-                spin_unlock(&dcache_lock);
-                cfs_spin_unlock(&ll_lookup_lock);
-                iput(inode);
-                CDEBUG(D_DENTRY, "alias 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, atomic_read(&de->d_count));
-                return dentry;
-        }
-
-        if (last_discon) {
-                CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
-                        "refc %d\n", last_discon, last_discon->d_inode,
-                        atomic_read(&last_discon->d_count));
-                dget_locked(last_discon);
-                lock_dentry(last_discon);
-                last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
-                unlock_dentry(last_discon);
-                spin_unlock(&dcache_lock);
-                cfs_spin_unlock(&ll_lookup_lock);
-                ll_dops_init(last_discon, 1);
-                d_rehash(de);
-                d_move(last_discon, de);
-                iput(inode);
-                return last_discon;
-        }
-        lock_dentry(de);
-        de->d_flags |= DCACHE_LUSTRE_INVALID;
-        unlock_dentry(de);
-        ll_d_add(de, inode);
-
-        spin_unlock(&dcache_lock);
-        cfs_spin_unlock(&ll_lookup_lock);
-
+       struct dentry *new;
+       int rc;
+
+       if (inode) {
+               new = ll_find_alias(inode, de);
+               if (new) {
+                       rc = ll_d_init(new);
+                       if (rc < 0) {
+                               dput(new);
+                               return ERR_PTR(rc);
+                       }
+                       d_move(new, de);
+                       iput(inode);
+                       CDEBUG(D_DENTRY,
+                              "Reuse dentry %p inode %p refc %d flags %#x\n",
+                             new, new->d_inode, d_count(new), new->d_flags);
+                       return new;
+               }
+       }
+       rc = ll_d_init(de);
+       if (rc < 0)
+               return ERR_PTR(rc);
+       d_add(de, inode);
+       CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
+              de, de->d_inode, d_count(de), de->d_flags);
         return de;
 }
 
-void ll_lookup_it_alias(struct dentry **de, struct inode *inode, __u32 bits)
-{
-        struct dentry *save = *de;
-        ENTRY;
-
-        ll_dops_init(*de, 1);
-        *de = ll_find_alias(inode, *de);
-        if (*de != save) {
-                struct ll_dentry_data *lld = ll_d2d(*de);
-
-                /* just make sure the ll_dentry_data is ready */
-                if (unlikely(lld == NULL))
-                        ll_dops_init(*de, 1);
-        }
-        /* we have lookup look - unhide dentry */
-        if (bits & MDS_INODELOCK_LOOKUP) {
-                lock_dentry(*de);
-                (*de)->d_flags &= ~DCACHE_LUSTRE_INVALID;
-                unlock_dentry(*de);
-        }
-        EXIT;
-}
-
 int ll_lookup_it_finish(struct ptlrpc_request *request,
                         struct lookup_intent *it, void *data)
 {
-        struct it_cb_data *icbd = data;
-        struct dentry **de = icbd->icbd_childp;
-        struct inode *parent = icbd->icbd_parent;
-        struct ll_sb_info *sbi = ll_i2sbi(parent);
-        struct inode *inode = NULL;
-        int rc;
-        ENTRY;
-
-        /* NB 1 request reference will be taken away by ll_intent_lock()
-         * when I return */
-        if (!it_disposition(it, DISP_LOOKUP_NEG)) {
-                __u32 bits;
-
-                rc = ll_prep_inode(&inode, request, (*de)->d_sb);
+       struct it_cb_data        *icbd = data;
+       struct dentry           **de = icbd->icbd_childp;
+       struct inode             *parent = icbd->icbd_parent;
+       struct inode             *inode = NULL;
+       __u64                     bits = 0;
+       int                       rc;
+       ENTRY;
+
+       /* NB 1 request reference will be taken away by ll_intent_lock()
+        * when I return */
+       CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
+              it->d.lustre.it_disposition);
+       if (!it_disposition(it, DISP_LOOKUP_NEG)) {
+                rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
                 if (rc)
                         RETURN(rc);
 
-                CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
-                       inode, inode->i_ino, inode->i_generation);
-                md_set_lock_data(sbi->ll_md_exp,
-                                 &it->d.lustre.it_lock_handle, inode, &bits);
-
-                if (icbd->bits != NULL)
-                        *icbd->bits = bits;
-                if (icbd->icbd_alias != NULL) {
-                        *icbd->icbd_alias = inode;
-                        RETURN(0);
-                }
+                ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
 
                 /* We used to query real size from OSTs here, but actually
                    this is not needed. For stat() calls size would be updated
@@ -497,34 +467,47 @@ int ll_lookup_it_finish(struct ptlrpc_request *request,
                    2.4 and
                    vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
                    Everybody else who needs correct file size would call
-                   cl_glimpse_size or some equivalent themselves anyway.
+                   ll_glimpse_size or some equivalent themselves anyway.
                    Also see bug 7198. */
-                ll_lookup_it_alias(de, inode, bits);
-        } else {
-                ll_dops_init(*de, 1);
-                /* Check that parent has UPDATE lock. If there is none, we
-                   cannot afford to hash this dentry (done by ll_d_add) as it
-                   might get picked up later when UPDATE lock will appear */
-                if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
-                        spin_lock(&dcache_lock);
-                        ll_d_add(*de, NULL);
-                        spin_unlock(&dcache_lock);
-                } else {
-                        /* negative lookup - and don't have update lock to
-                         * parent */
-                        lock_dentry(*de);
-                        (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
-                        unlock_dentry(*de);
-
-                        (*de)->d_inode = NULL;
-                        /* We do not want to hash the dentry if don`t have a
-                         * lock, but if this dentry is later used in d_move,
-                         * we'd hit uninitialised list head d_hash, so we just
-                         * do this to init d_hash field but leave dentry
-                         * unhashed. (bug 10796). */
-                        d_rehash(*de);
-                        d_drop(*de);
-                }
+       }
+
+       /* Only hash *de if it is unhashed (new dentry).
+        * Atoimc_open may passin hashed dentries for open.
+        */
+       if (d_unhashed(*de)) {
+               struct dentry *alias;
+
+               alias = ll_splice_alias(inode, *de);
+               if (IS_ERR(alias))
+                       RETURN(PTR_ERR(alias));
+               *de = alias;
+       } else if (!it_disposition(it, DISP_LOOKUP_NEG)  &&
+                  !it_disposition(it, DISP_OPEN_CREATE)) {
+               /* With DISP_OPEN_CREATE dentry will
+                  instantiated in ll_create_it. */
+               LASSERT((*de)->d_inode == NULL);
+               d_instantiate(*de, inode);
+       }
+
+       if (!it_disposition(it, DISP_LOOKUP_NEG)) {
+               /* we have lookup look - unhide dentry */
+               if (bits & MDS_INODELOCK_LOOKUP)
+                       d_lustre_revalidate(*de);
+       } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
+               /* If file created on server, don't depend on parent UPDATE
+                * lock to unhide it. It is left hidden and next lookup can
+                * find it in ll_splice_alias.
+                */
+               /* Check that parent has UPDATE lock. */
+               struct lookup_intent parent_it = {
+                                       .it_op = IT_GETATTR,
+                                       .d.lustre.it_lock_handle = 0 };
+
+               if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
+                                      &ll_i2info(parent)->lli_fid, NULL)) {
+                       d_lustre_revalidate(*de);
+                       ll_intent_release(&parent_it);
+               }
         }
 
         RETURN(0);
@@ -539,49 +522,37 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
         struct md_op_data *op_data;
         struct it_cb_data icbd;
         __u32 opc;
-        int rc, first = 0;
+        int rc;
         ENTRY;
 
         if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
                 RETURN(ERR_PTR(-ENAMETOOLONG));
 
-        CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
-               dentry->d_name.len, dentry->d_name.name, parent->i_ino,
-               parent->i_generation, parent, LL_IT2STR(it));
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
+              dentry->d_name.len, dentry->d_name.name,
+              PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
 
         if (d_mountpoint(dentry))
                 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
 
         ll_frob_intent(&it, &lookup_it);
 
-        /* As do_lookup is called before follow_mount, root dentry may be left
-         * not valid, revalidate it here. */
-        if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
-            (it->it_op & (IT_OPEN | IT_CREAT))) {
-                rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
-                if (rc)
-                        RETURN(ERR_PTR(rc));
-        }
-
         if (it->it_op == IT_GETATTR) {
-                first = ll_statahead_enter(parent, &dentry, 1);
-                if (first >= 0) {
-                        ll_statahead_exit(parent, dentry, first);
-                        if (first == 1)
-                                RETURN(retval = dentry);
+                rc = ll_statahead_enter(parent, &dentry, 0);
+                if (rc == 1) {
+                        if (dentry == save)
+                                GOTO(out, retval = NULL);
+                        GOTO(out, retval = dentry);
                 }
         }
 
         icbd.icbd_childp = &dentry;
         icbd.icbd_parent = parent;
-        icbd.icbd_alias  = NULL;
-        icbd.bits        = NULL;
 
-        if (it->it_op & IT_CREAT ||
-            (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
-                opc = LUSTRE_OPC_CREATE;
-        else
-                opc = LUSTRE_OPC_ANY;
+       if (it->it_op & IT_CREAT)
+               opc = LUSTRE_OPC_CREATE;
+       else
+               opc = LUSTRE_OPC_ANY;
 
         op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
                                      dentry->d_name.len, lookup_flags, opc,
@@ -589,7 +560,9 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
         if (IS_ERR(op_data))
                 RETURN((void *)op_data);
 
-        it->it_create_mode &= ~current->fs->umask;
+        /* enforce umask if acl disabled or MDS doesn't support umask */
+        if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
+               it->it_create_mode &= ~current_umask();
 
         rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
                             lookup_flags, &req, ll_md_blocking_ast, 0);
@@ -603,9 +576,6 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
                 GOTO(out, retval = ERR_PTR(rc));
         }
 
-        if (first == -EEXIST)
-                ll_statahead_mark(parent, dentry);
-
         if ((it->it_op & IT_OPEN) && dentry->d_inode &&
             !S_ISREG(dentry->d_inode->i_mode) &&
             !S_ISDIR(dentry->d_inode->i_mode)) {
@@ -620,52 +590,142 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
  out:
         if (req)
                 ptlrpc_req_finished(req);
+        if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
+                ll_statahead_mark(parent, dentry);
         return retval;
 }
 
-#ifdef HAVE_VFS_INTENT_PATCHES
+#ifdef HAVE_IOP_ATOMIC_OPEN
 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
-                                   struct nameidata *nd)
+                                  unsigned int flags)
 {
-        struct dentry *de;
-        ENTRY;
+       struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
+       struct dentry *de;
 
-        if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
-                de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
-        else
-                de = ll_lookup_it(parent, dentry, NULL, 0);
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
+              dentry->d_name.len, dentry->d_name.name,
+              PFID(ll_inode2fid(parent)), parent, flags);
 
-        RETURN(de);
+       /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
+       if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
+               return NULL;
+
+       if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
+               itp = NULL;
+       else
+               itp = &it;
+       de = ll_lookup_it(parent, dentry, itp, 0);
+
+       if (itp != NULL)
+               ll_intent_release(itp);
+
+       return de;
 }
-#else
+
+/*
+ * For cached negative dentry and new dentry, handle lookup/create/open
+ * together.
+ */
+static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
+                         struct file *file, unsigned open_flags,
+                         umode_t mode, int *opened)
+{
+       struct lookup_intent *it;
+       struct dentry *de;
+       long long lookup_flags = LOOKUP_OPEN;
+       int rc = 0;
+       ENTRY;
+
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
+                          "open_flags %x, mode %x opened %d\n",
+              dentry->d_name.len, dentry->d_name.name,
+              PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
+
+       OBD_ALLOC(it, sizeof(*it));
+       if (!it)
+               RETURN(-ENOMEM);
+
+       it->it_op = IT_OPEN;
+       if (open_flags & O_CREAT) {
+               it->it_op |= IT_CREAT;
+               lookup_flags |= LOOKUP_CREATE;
+       }
+       it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
+       it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
+
+       /* Dentry added to dcache tree in ll_lookup_it */
+       de = ll_lookup_it(dir, dentry, it, lookup_flags);
+       if (IS_ERR(de))
+               rc = PTR_ERR(de);
+       else if (de != NULL)
+               dentry = de;
+
+       if (!rc) {
+               if (it_disposition(it, DISP_OPEN_CREATE)) {
+                       /* Dentry instantiated in ll_create_it. */
+                       rc = ll_create_it(dir, dentry, mode, it);
+                       if (rc) {
+                               /* We dget in ll_splice_alias. */
+                               if (de != NULL)
+                                       dput(de);
+                               goto out_release;
+                       }
+
+                       *opened |= FILE_CREATED;
+               }
+               if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
+                       /* Open dentry. */
+                       if (S_ISFIFO(dentry->d_inode->i_mode)) {
+                               /* We cannot call open here as it would
+                                * deadlock.
+                                */
+                               if (it_disposition(it, DISP_ENQ_OPEN_REF))
+                                       ptlrpc_req_finished(
+                                                      (struct ptlrpc_request *)
+                                                         it->d.lustre.it_data);
+                               rc = finish_no_open(file, de);
+                       } else {
+                               file->private_data = it;
+                               rc = finish_open(file, dentry, NULL, opened);
+                               /* We dget in ll_splice_alias. finish_open takes
+                                * care of dget for fd open.
+                                */
+                               if (de != NULL)
+                                       dput(de);
+                       }
+               } else {
+                       rc = finish_no_open(file, de);
+               }
+       }
+
+out_release:
+       ll_intent_release(it);
+       OBD_FREE(it, sizeof(*it));
+
+       RETURN(rc);
+}
+
+#else /* !HAVE_IOP_ATOMIC_OPEN */
 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
                                         int lookup_flags)
 {
-        struct lookup_intent *it;
-
-        OBD_ALLOC(it, sizeof(*it));
-        if (!it)
-                return ERR_PTR(-ENOMEM);
-
-        if (lookup_flags & LOOKUP_OPEN) {
-                it->it_op = IT_OPEN;
-                if (lookup_flags & LOOKUP_CREATE)
-                        it->it_op |= IT_CREAT;
-                it->it_create_mode = oit->create_mode;
-                it->it_flags = oit->flags;
-        } else {
-                it->it_op = IT_GETATTR;
-        }
-
-#ifndef HAVE_FILE_IN_STRUCT_INTENT
-                /* Since there is no way to pass our intent to ll_file_open,
-                 * just check the file is there. Actual open will be done
-                 * in ll_file_open */
-                if (it->it_op & IT_OPEN)
-                        it->it_op = IT_LOOKUP;
-#endif
-
-        return it;
+       struct lookup_intent *it;
+
+       OBD_ALLOC(it, sizeof(*it));
+       if (!it)
+               return ERR_PTR(-ENOMEM);
+
+       if (lookup_flags & LOOKUP_OPEN) {
+               it->it_op = IT_OPEN;
+               if (lookup_flags & LOOKUP_CREATE)
+                       it->it_op |= IT_CREAT;
+               it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
+               it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
+       } else {
+               it->it_op = IT_GETATTR;
+       }
+
+       return it;
 }
 
 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
@@ -677,24 +737,14 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
         if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
                 struct lookup_intent *it;
 
-#if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
-                /* Did we came here from failed revalidate just to propagate
-                 * its error? */
-                if (nd->flags & LOOKUP_OPEN)
-                        if (IS_ERR(nd->intent.open.file))
-                                RETURN((struct dentry *)nd->intent.open.file);
-#endif
-
                 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
                         it = ll_d2d(dentry)->lld_it;
                         ll_d2d(dentry)->lld_it = NULL;
                 } else {
-                        if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
-                                /* We are sure this is new dentry, so we need to create
-                                   our private data and set the dentry ops */ 
-                                ll_dops_init(dentry, 1);
+                       if ((nd->flags & LOOKUP_CREATE) &&
+                           !(nd->flags & LOOKUP_OPEN))
                                 RETURN(NULL);
-                        }
+
                         it = ll_convert_intent(&nd->intent.open, nd->flags);
                         if (IS_ERR(it))
                                 RETURN((struct dentry *)it);
@@ -706,7 +756,6 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
                         if (dentry->d_inode &&
                             it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
-#ifdef HAVE_FILE_IN_STRUCT_INTENT
                                 if (S_ISFIFO(dentry->d_inode->i_mode)) {
                                         // We cannot call open here as it would
                                         // deadlock.
@@ -714,31 +763,18 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
                                                        (struct ptlrpc_request *)
                                                           it->d.lustre.it_data);
                                 } else {
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
-/* 2.6.1[456] have a bug in open_namei() that forgets to check
- * nd->intent.open.file for error, so we need to return it as lookup's result
- * instead */
-                                        struct file *filp;
-                                        nd->intent.open.file->private_data = it;
-                                        filp =lookup_instantiate_filp(nd,dentry,
-                                                                      NULL);
-                                        if (IS_ERR(filp)) {
-                                                if (de)
-                                                        dput(de);
-                                                de = (struct dentry *) filp;
-                                        }
-#else
-                                        nd->intent.open.file->private_data = it;
-                                        (void)lookup_instantiate_filp(nd,dentry,
-                                                                      NULL);
-#endif
-
+                                       struct file *filp;
+
+                                       nd->intent.open.file->private_data = it;
+                                       filp = lookup_instantiate_filp(nd,
+                                                                      dentry,
+                                                                      NULL);
+                                       if (IS_ERR(filp)) {
+                                               if (de)
+                                                       dput(de);
+                                               de = (struct dentry *)filp;
+                                       }
                                 }
-#else /* HAVE_FILE_IN_STRUCT_INTENT */
-                                /* Release open handle as we have no way to
-                                 * pass it to ll_file_open */
-                                ll_release_openhandle(dentry, it);
-#endif /* HAVE_FILE_IN_STRUCT_INTENT */
                         } else if (it_disposition(it, DISP_OPEN_CREATE)) {
                                 // XXX This can only reliably work on assumption
                                 // that there are NO hashed negative dentries.
@@ -760,7 +796,7 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
 
         RETURN(de);
 }
-#endif
+#endif /* HAVE_IOP_ATOMIC_OPEN */
 
 /* We depend on "mode" being set with the proper file type/umask by now */
 static struct inode *ll_create_node(struct inode *dir, const char *name,
@@ -779,19 +815,18 @@ static struct inode *ll_create_node(struct inode *dir, const char *name,
         LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
         request = it->d.lustre.it_data;
         it_clear_disposition(it, DISP_ENQ_CREATE_REF);
-        rc = ll_prep_inode(&inode, request, dir->i_sb);
+        rc = ll_prep_inode(&inode, request, dir->i_sb, it);
         if (rc)
                 GOTO(out, inode = ERR_PTR(rc));
 
-        LASSERT(list_empty(&inode->i_dentry));
+       LASSERT(ll_d_hlist_empty(&inode->i_dentry));
 
         /* We asked for a lock on the directory, but were granted a
          * lock on the inode.  Since we finally have an inode pointer,
          * stuff it in the lock. */
-        CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
-               inode, inode->i_ino, inode->i_generation);
-        md_set_lock_data(sbi->ll_md_exp,
-                         &it->d.lustre.it_lock_handle, inode, NULL);
+       CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
+              PFID(ll_inode2fid(inode)), inode);
+        ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
         EXIT;
  out:
         ptlrpc_req_finished(request);
@@ -813,35 +848,30 @@ static struct inode *ll_create_node(struct inode *dir, const char *name,
  * with d_instantiate().
  */
 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
-                        struct lookup_intent *it)
+                       struct lookup_intent *it)
 {
-        struct inode *inode;
-        int rc = 0;
-        ENTRY;
+       struct inode *inode;
+       int rc = 0;
+       ENTRY;
 
-        CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
-               dentry->d_name.len, dentry->d_name.name, dir->i_ino,
-               dir->i_generation, dir, LL_IT2STR(it));
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
+              dentry->d_name.len, dentry->d_name.name,
+              PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
 
-        rc = it_open_error(DISP_OPEN_CREATE, it);
-        if (rc)
-                RETURN(rc);
+       rc = it_open_error(DISP_OPEN_CREATE, it);
+       if (rc)
+               RETURN(rc);
 
-        inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
-                               NULL, 0, mode, 0, it);
-        if (IS_ERR(inode)) {
-                RETURN(PTR_ERR(inode));
-        }
+       inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
+                              NULL, 0, mode, 0, it);
+       if (IS_ERR(inode))
+               RETURN(PTR_ERR(inode));
 
-        d_instantiate(dentry, inode);
-        /* Negative dentry may be unhashed if parent does not have UPDATE lock,
-         * but some callers, e.g. do_coredump, expect dentry to be hashed after
-         * successful create. Hash it here. */
-        spin_lock(&dcache_lock);
-        if (d_unhashed(dentry))
-                d_rehash_cond(dentry, 0);
-        spin_unlock(&dcache_lock);
-        RETURN(0);
+       if (filename_is_volatile(dentry->d_name.name, dentry->d_name.len, NULL))
+               ll_i2info(inode)->lli_volatile = true;
+
+       d_instantiate(dentry, inode);
+       RETURN(0);
 }
 
 static void ll_update_times(struct ptlrpc_request *request,
@@ -853,8 +883,9 @@ static void ll_update_times(struct ptlrpc_request *request,
         LASSERT(body);
         if (body->valid & OBD_MD_FLMTIME &&
             body->mtime > LTIME_S(inode->i_mtime)) {
-                CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
-                       inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
+               CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
+                               "\n", PFID(ll_inode2fid(inode)),
+                               LTIME_S(inode->i_mtime), body->mtime);
                 LTIME_S(inode->i_mtime) = body->mtime;
         }
         if (body->valid & OBD_MD_FLCTIME &&
@@ -882,24 +913,23 @@ static int ll_new_node(struct inode *dir, struct qstr *name,
         if (IS_ERR(op_data))
                 GOTO(err_exit, err = PTR_ERR(op_data));
 
-        err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
-                        cfs_curproc_fsuid(), cfs_curproc_fsgid(),
-                        cfs_curproc_cap_pack(), rdev, &request);
-        ll_finish_md_op_data(op_data);
+       err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
+                       current_fsuid(), current_fsgid(),
+                       cfs_curproc_cap_pack(), rdev, &request);
+       ll_finish_md_op_data(op_data);
         if (err)
                 GOTO(err_exit, err);
 
         ll_update_times(request, dir);
 
         if (dchild) {
-                err = ll_prep_inode(&inode, request, dchild->d_sb);
+                err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
                 if (err)
                      GOTO(err_exit, err);
 
-                d_drop(dchild);
                 d_instantiate(dchild, inode);
-                EXIT;
         }
+        EXIT;
 err_exit:
         ptlrpc_req_finished(request);
 
@@ -912,11 +942,12 @@ static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
         int err;
         ENTRY;
 
-        CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
-               name->len, name->name, dir->i_ino, dir->i_generation, dir,
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
+              name->len, name->name, PFID(ll_inode2fid(dir)), dir,
                mode, rdev);
 
-        mode &= ~current->fs->umask;
+       if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
+               mode &= ~current_umask();
 
         switch (mode & S_IFMT) {
         case 0:
@@ -935,20 +966,51 @@ static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
         default:
                 err = -EINVAL;
         }
+
+        if (!err)
+                ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
+
         RETURN(err);
 }
 
-#ifndef HAVE_VFS_INTENT_PATCHES
+#ifdef HAVE_IOP_ATOMIC_OPEN
+/*
+ * Plain create. Intent create is handled in atomic_open.
+ */
+static int ll_create_nd(struct inode *dir, struct dentry *dentry,
+                       umode_t mode, bool want_excl)
+{
+       int rc;
+
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
+                          "flags=%u, excl=%d\n", dentry->d_name.len,
+              dentry->d_name.name, PFID(ll_inode2fid(dir)),
+              dir, mode, want_excl);
+
+       rc = ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
+
+       ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
+
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
+              dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
+
+       return rc;
+}
+#else /* !HAVE_IOP_ATOMIC_OPEN */
 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
-                        int mode, struct nameidata *nd)
+                       ll_umode_t mode, struct nameidata *nd)
 {
-        struct lookup_intent *it = ll_d2d(dentry)->lld_it;
+       struct ll_dentry_data *lld = ll_d2d(dentry);
+       struct lookup_intent *it = NULL;
         int rc;
 
+       if (lld != NULL)
+               it = lld->lld_it;
+
         if (!it)
                 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
 
-        ll_d2d(dentry)->lld_it = NULL;
+       lld->lld_it = NULL;
 
         /* Was there an error? Propagate it! */
         if (it->d.lustre.it_status) {
@@ -957,32 +1019,25 @@ static int ll_create_nd(struct inode *dir, struct dentry *dentry,
         }
 
         rc = ll_create_it(dir, dentry, mode, it);
-#ifdef HAVE_FILE_IN_STRUCT_INTENT
         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
-                nd->intent.open.file->private_data = it;
-                lookup_instantiate_filp(nd, dentry, NULL);
+               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);
         }
-#else
-        ll_release_openhandle(dentry,it);
-#endif
 
 out:
         ll_intent_release(it);
         OBD_FREE(it, sizeof(*it));
 
+        if (!rc)
+                ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
+
         return rc;
 }
-#else
-static int ll_create_nd(struct inode *dir, struct dentry *dentry,
-                        int mode, struct nameidata *nd)
-{
-        if (!nd || !nd->intent.d.lustre.it_disposition)
-                /* No saved request? Just mknod the file */
-                return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
-
-        return ll_create_it(dir, dentry, mode, &nd->intent);
-}
-#endif
+#endif /* HAVE_IOP_ATOMIC_OPEN */
 
 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
                               const char *tgt, struct dentry *dchild)
@@ -990,12 +1045,16 @@ static int ll_symlink_generic(struct inode *dir, struct qstr *name,
         int err;
         ENTRY;
 
-        CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
-               name->len, name->name, dir->i_ino, dir->i_generation,
-               dir, 3000, tgt);
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
+              name->len, name->name, PFID(ll_inode2fid(dir)),
+              dir, 3000, tgt);
 
         err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
                           0, dchild, LUSTRE_OPC_SYMLINK);
+
+        if (!err)
+                ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
+
         RETURN(err);
 }
 
@@ -1008,10 +1067,9 @@ static int ll_link_generic(struct inode *src,  struct inode *dir,
         int err;
 
         ENTRY;
-        CDEBUG(D_VFSTRACE,
-               "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
-               src->i_ino, src->i_generation, src, dir->i_ino,
-               dir->i_generation, dir, name->len, name->name);
+       CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID
+              "(%p), target=%.*s\n", PFID(ll_inode2fid(src)), src,
+              PFID(ll_inode2fid(dir)), dir, name->len, name->name);
 
         op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
                                      0, LUSTRE_OPC_ANY, NULL);
@@ -1022,10 +1080,9 @@ static int ll_link_generic(struct inode *src,  struct inode *dir,
         ll_finish_md_op_data(op_data);
         if (err)
                 GOTO(out, err);
-        if (dchild)
-                d_drop(dchild);
 
         ll_update_times(request, dir);
+        ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
         EXIT;
 out:
         ptlrpc_req_finished(request);
@@ -1039,29 +1096,18 @@ static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
         int err;
         ENTRY;
 
-        CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
-               name->len, name->name, dir->i_ino, dir->i_generation, dir);
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
+              name->len, name->name, PFID(ll_inode2fid(dir)), dir);
 
-        mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
-        err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
+       if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
+               mode &= ~current_umask();
+       mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
+       err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
 
-        RETURN(err);
-}
+        if (!err)
+                ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
 
-/* Try to find the child dentry by its name.
-   If found, put the result fid into @fid. */
-static void ll_get_child_fid(struct inode * dir, struct qstr *name,
-                             struct lu_fid *fid)
-{
-        struct dentry *parent, *child;
-
-        parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
-        child = d_lookup(parent, name);
-        if (child) {
-                if (child->d_inode)
-                        *fid = *ll_inode2fid(child->d_inode);
-                dput(child);
-        }
+        RETURN(err);
 }
 
 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
@@ -1072,8 +1118,8 @@ static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
         int rc;
         ENTRY;
 
-        CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
-               name->len, name->name, dir->i_ino, dir->i_generation, dir);
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
+              name->len, name->name, PFID(ll_inode2fid(dir)), dir);
 
         if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
                 RETURN(-EBUSY);
@@ -1083,15 +1129,49 @@ static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
         if (IS_ERR(op_data))
                 RETURN(PTR_ERR(op_data));
 
-        ll_get_child_fid(dir, name, &op_data->op_fid3);
+       if (dchild != NULL && dchild->d_inode != NULL)
+               op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
+       op_data->op_fid2 = op_data->op_fid3;
         rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
         ll_finish_md_op_data(op_data);
-        if (rc == 0)
+        if (rc == 0) {
                 ll_update_times(request, dir);
+                ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
+        }
+
         ptlrpc_req_finished(request);
         RETURN(rc);
 }
 
+/**
+ * Remove dir entry
+ **/
+int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
+{
+       struct ptlrpc_request *request = NULL;
+       struct md_op_data *op_data;
+       int rc;
+       ENTRY;
+
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
+              namelen, name, PFID(ll_inode2fid(dir)), dir);
+
+       op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
+                                    S_IFDIR, LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
+       op_data->op_cli_flags |= CLI_RM_ENTRY;
+       rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
+       ll_finish_md_op_data(op_data);
+       if (rc == 0) {
+               ll_update_times(request, dir);
+               ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
+       }
+
+       ptlrpc_req_finished(request);
+       RETURN(rc);
+}
+
 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
 {
         struct mdt_body *body;
@@ -1132,8 +1212,7 @@ int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
         if (oa == NULL)
                 GOTO(out_free_memmd, rc = -ENOMEM);
 
-        oa->o_id = lsm->lsm_object_id;
-        oa->o_seq = lsm->lsm_object_seq;
+       oa->o_oi = lsm->lsm_oi;
         oa->o_mode = body->mode & S_IFMT;
         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
 
@@ -1156,16 +1235,17 @@ int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
                         GOTO(out_free_memmd, rc);
         }
 
-        rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
-        capa_put(oc);
-        OBDO_FREE(oa);
-        if (rc)
-                CERROR("obd destroy objid "LPX64" error %d\n",
-                       lsm->lsm_object_id, rc);
- out_free_memmd:
-        obd_free_memmd(ll_i2dtexp(dir), &lsm);
- out:
-        return rc;
+       rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
+                        ll_i2mdexp(dir), oc);
+       capa_put(oc);
+       if (rc)
+               CERROR("obd destroy objid "DOSTID" error %d\n",
+                      POSTID(&lsm->lsm_oi), rc);
+out_free_memmd:
+       obd_free_memmd(ll_i2dtexp(dir), &lsm);
+       OBDO_FREE(oa);
+out:
+       return rc;
 }
 
 /* ll_unlink_generic() doesn't update the inode with the new link count.
@@ -1179,8 +1259,8 @@ static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
         struct md_op_data *op_data;
         int rc;
         ENTRY;
-        CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
-               name->len, name->name, dir->i_ino, dir->i_generation, dir);
+       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
+              name->len, name->name, PFID(ll_inode2fid(dir)), dir);
 
         /*
          * XXX: unlink bind mountpoint maybe call to here,
@@ -1194,13 +1274,17 @@ static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
         if (IS_ERR(op_data))
                 RETURN(PTR_ERR(op_data));
 
-        ll_get_child_fid(dir, name, &op_data->op_fid3);
-        rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
-        ll_finish_md_op_data(op_data);
-        if (rc)
-                GOTO(out, rc);
+       if (dchild != NULL && dchild->d_inode != NULL)
+               op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
+
+       op_data->op_fid2 = op_data->op_fid3;
+       rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
+       ll_finish_md_op_data(op_data);
+       if (rc)
+               GOTO(out, rc);
 
         ll_update_times(request, dir);
+        ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
 
         rc = ll_objects_destroy(request, dir);
  out:
@@ -1218,10 +1302,11 @@ static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
         struct md_op_data *op_data;
         int err;
         ENTRY;
-        CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
-               "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
-               src->i_ino, src->i_generation, src, tgt_name->len,
-               tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
+       CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
+              "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
+              src_name->len, src_name->name,
+              PFID(ll_inode2fid(src)), src, tgt_name->len,
+              tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
 
         if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
             ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
@@ -1232,8 +1317,11 @@ static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
         if (IS_ERR(op_data))
                 RETURN(PTR_ERR(op_data));
 
-        ll_get_child_fid(src, src_name, &op_data->op_fid3);
-        ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
+       if (src_dchild != NULL && src_dchild->d_inode != NULL)
+               op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
+       if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
+               op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
+
         err = md_rename(sbi->ll_md_exp, op_data,
                         src_name->name, src_name->len,
                         tgt_name->name, tgt_name->len, &request);
@@ -1241,6 +1329,7 @@ static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
         if (!err) {
                 ll_update_times(request, src);
                 ll_update_times(request, tgt);
+                ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
                 err = ll_objects_destroy(request, src);
         }
 
@@ -1249,73 +1338,41 @@ static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
         RETURN(err);
 }
 
-#ifdef HAVE_VFS_INTENT_PATCHES
-static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
-{
-        return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
-}
-static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
-{
-        return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
-                                 NULL, &srcnd->last,
-                                 tgtnd->dentry->d_inode, tgtnd->dentry,
-                                 NULL, &tgtnd->last);
-}
-static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
-{
-        return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
-                               &tgtnd->last, NULL);
-}
-static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
-{
-        return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
-}
-static int ll_rmdir_raw(struct nameidata *nd)
-{
-        return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
-                                &nd->last);
-}
-static int ll_mkdir_raw(struct nameidata *nd, int mode)
-{
-        return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
-}
-static int ll_unlink_raw(struct nameidata *nd)
-{
-        return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
-                                 &nd->last);
-}
-#endif
-
-static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
-                    ll_dev_t rdev)
+static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
+                   dev_t rdev)
 {
-        return ll_mknod_generic(dir, &dchild->d_name, mode,
-                                old_encode_dev(rdev), dchild);
+       return ll_mknod_generic(dir, &dchild->d_name, mode,
+                               old_encode_dev(rdev), dchild);
 }
 
 static int ll_unlink(struct inode * dir, struct dentry *dentry)
 {
         return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
 }
-static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
+
+static int ll_mkdir(struct inode *dir, struct dentry *dentry, ll_umode_t mode)
 {
         return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
 }
+
 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
 {
         return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
 }
+
 static int ll_symlink(struct inode *dir, struct dentry *dentry,
                       const char *oldname)
 {
         return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
 }
+
 static int ll_link(struct dentry *old_dentry, struct inode *dir,
                    struct dentry *new_dentry)
 {
         return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
                                new_dentry);
 }
+
 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
                      struct inode *new_dir, struct dentry *new_dentry)
 {
@@ -1324,55 +1381,46 @@ static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
                                  old_dentry, &old_dentry->d_name,
                                  new_dir, NULL, new_dentry,
                                  &new_dentry->d_name);
-        if (!err) {
-#ifndef HAVE_FS_RENAME_DOES_D_MOVE
-                if (!S_ISDIR(old_dentry->d_inode->i_mode))
-#endif
+       if (!err)
                         d_move(old_dentry, new_dentry);
-        }
         return err;
 }
 
 struct inode_operations ll_dir_inode_operations = {
-#ifdef HAVE_VFS_INTENT_PATCHES
-        .link_raw           = ll_link_raw,
-        .unlink_raw         = ll_unlink_raw,
-        .symlink_raw        = ll_symlink_raw,
-        .mkdir_raw          = ll_mkdir_raw,
-        .rmdir_raw          = ll_rmdir_raw,
-        .mknod_raw          = ll_mknod_raw,
-        .rename_raw         = ll_rename_raw,
-        .setattr            = ll_setattr,
-        .setattr_raw        = ll_setattr_raw,
+       .mknod              = ll_mknod,
+#ifdef HAVE_IOP_ATOMIC_OPEN
+       .atomic_open        = ll_atomic_open,
+#endif
+       .lookup             = ll_lookup_nd,
+       .create             = ll_create_nd,
+       /* We need all these non-raw things for NFSD, to not patch it. */
+       .unlink             = ll_unlink,
+       .mkdir              = ll_mkdir,
+       .rmdir              = ll_rmdir,
+       .symlink            = ll_symlink,
+       .link               = ll_link,
+       .rename             = ll_rename,
+       .setattr            = ll_setattr,
+       .getattr            = ll_getattr,
+       .permission         = ll_inode_permission,
+       .setxattr           = ll_setxattr,
+       .getxattr           = ll_getxattr,
+       .listxattr          = ll_listxattr,
+       .removexattr        = ll_removexattr,
+#ifdef HAVE_IOP_GET_ACL
+       .get_acl            = ll_get_acl,
 #endif
-        .mknod              = ll_mknod,
-        .lookup             = ll_lookup_nd,
-        .create             = ll_create_nd,
-        /* We need all these non-raw things for NFSD, to not patch it. */
-        .unlink             = ll_unlink,
-        .mkdir              = ll_mkdir,
-        .rmdir              = ll_rmdir,
-        .symlink            = ll_symlink,
-        .link               = ll_link,
-        .rename             = ll_rename,
-        .setattr            = ll_setattr,
-        .getattr            = ll_getattr,
-        .permission         = ll_inode_permission,
-        .setxattr           = ll_setxattr,
-        .getxattr           = ll_getxattr,
-        .listxattr          = ll_listxattr,
-        .removexattr        = ll_removexattr,
 };
 
 struct inode_operations ll_special_inode_operations = {
-#ifdef HAVE_VFS_INTENT_PATCHES
-        .setattr_raw    = ll_setattr_raw,
+       .setattr        = ll_setattr,
+       .getattr        = ll_getattr,
+       .permission     = ll_inode_permission,
+       .setxattr       = ll_setxattr,
+       .getxattr       = ll_getxattr,
+       .listxattr      = ll_listxattr,
+       .removexattr    = ll_removexattr,
+#ifdef HAVE_IOP_GET_ACL
+       .get_acl            = ll_get_acl,
 #endif
-        .setattr        = ll_setattr,
-        .getattr        = ll_getattr,
-        .permission     = ll_inode_permission,
-        .setxattr       = ll_setxattr,
-        .getxattr       = ll_getxattr,
-        .listxattr      = ll_listxattr,
-        .removexattr    = ll_removexattr,
 };