Whamcloud - gitweb
LU-6301 llite: cleanup open handle for client open failure
[fs/lustre-release.git] / lustre / llite / namei.c
index aaf864b..f5288a9 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, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2014, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 
 #include <obd_support.h>
 #include <lustre_fid.h>
-#include <lustre_lite.h>
 #include <lustre_dlm.h>
 #include <lustre_ver.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.
- */
-static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
-                           struct qstr *name)
-{
-        int mounted = 0;
-
-        if (unlikely(dchild)) {
-                mounted = d_mountpoint(dchild);
-        } else if (dparent) {
-                dchild = d_lookup(dparent, name);
-                if (dchild) {
-                        mounted = d_mountpoint(dchild);
-                        dput(dchild);
-                }
-        }
-        return mounted;
-}
-
-int ll_unlock(__u32 mode, struct lustre_handle *lockh)
-{
-        ENTRY;
-
-        ldlm_lock_decref(lockh, mode);
-
-        RETURN(0);
-}
-
+static int ll_create_it(struct inode *dir, struct dentry *dentry,
+                       struct lookup_intent *it);
 
 /* called from iget5_locked->find_inode() under inode_lock spinlock */
 static int ll_test_inode(struct inode *inode, void *opaque)
 {
-        struct ll_inode_info *lli = ll_i2info(inode);
-        struct lustre_md     *md = opaque;
+       struct ll_inode_info    *lli = ll_i2info(inode);
+       struct lustre_md        *md = opaque;
 
-        if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
-                CERROR("MDS body missing FID\n");
-                return 0;
-        }
+       if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
+               CERROR("MDS body missing FID\n");
+               return 0;
+       }
 
-        if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
-                return 0;
+       if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
+               return 0;
 
-        return 1;
+       return 1;
 }
 
 static int ll_set_inode(struct inode *inode, void *opaque)
 {
-        struct ll_inode_info *lli = ll_i2info(inode);
-        struct mdt_body *body = ((struct lustre_md *)opaque)->body;
+       struct ll_inode_info *lli = ll_i2info(inode);
+       struct mdt_body *body = ((struct lustre_md *)opaque)->body;
 
-        if (unlikely(!(body->valid & OBD_MD_FLID))) {
-                CERROR("MDS body missing FID\n");
-                return -EINVAL;
-        }
+       if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
+               CERROR("MDS body missing FID\n");
+               return -EINVAL;
+       }
 
-        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;
-        }
+       lli->lli_fid = body->mbo_fid1;
+       if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
+               CERROR("Can not initialize inode "DFID" without object type: "
+                      "valid = "LPX64"\n",
+                      PFID(&lli->lli_fid), body->mbo_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;
-        }
+       inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_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);
+       ll_lli_init(lli);
 
-        return 0;
+       return 0;
 }
 
 
-/*
- * Get an inode by inode number (already instantiated by the intent lookup).
- * Returns inode or NULL
+/**
+ * Get an inode by inode number(@hash), which is already instantiated by
+ * the intent lookup).
  */
 struct inode *ll_iget(struct super_block *sb, ino_t hash,
                       struct lustre_md *md)
 {
-        struct inode         *inode;
-        ENTRY;
+       struct inode    *inode;
+       int             rc = 0;
+
+       ENTRY;
 
         LASSERT(hash != 0);
         inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
+       if (inode == NULL)
+               RETURN(ERR_PTR(-ENOMEM));
+
+       if (inode->i_state & I_NEW) {
+               rc = ll_read_inode2(inode, md);
+               if (rc == 0 && 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) {
+                       make_bad_inode(inode);
+                       unlock_new_inode(inode);
+                       iput(inode);
+                       inode = ERR_PTR(rc);
+               } else {
+                       unlock_new_inode(inode);
+               }
+       } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
+               rc = ll_update_inode(inode, md);
+               CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
+                      PFID(&md->body->mbo_fid1), inode, rc);
+               if (rc != 0) {
+                       iput(inode);
+                       inode = ERR_PTR(rc);
+               }
+       }
 
-        if (inode) {
-                if (inode->i_state & I_NEW) {
-                        int rc = 0;
-
-                        ll_read_inode2(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) {
-                                make_bad_inode(inode);
-                                unlock_new_inode(inode);
-                                iput(inode);
-                                inode = ERR_PTR(rc);
-                        } 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));
-        }
         RETURN(inode);
 }
 
@@ -195,121 +170,166 @@ static void ll_invalidate_negative_children(struct inode *dir)
        ll_unlock_dcache(dir);
 }
 
+int ll_test_inode_by_fid(struct inode *inode, void *opaque)
+{
+       return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
+}
+
 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;
+       struct lustre_handle lockh;
+       int rc;
+       ENTRY;
 
-        switch (flag) {
-        case LDLM_CB_BLOCKING:
-                ldlm_lock2handle(lock, &lockh);
+       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: %d\n", rc);
-                        RETURN(rc);
-                }
-                break;
-        case LDLM_CB_CANCELING: {
+               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);
-                struct ll_inode_info *lli;
-                __u64 bits = lock->l_policy_data.l_inodebits.bits;
-                struct lu_fid *fid;
-                ldlm_mode_t mode = lock->l_req_mode;
+               __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);
 
-               /* Invalidate all dentries associated with this inode */
                if (inode == NULL)
                        break;
 
-               LASSERT(lock->l_flags & LDLM_FL_CANCELING);
+               /* 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)
+               if (bits & MDS_INODELOCK_XATTR) {
+                       if (S_ISDIR(inode->i_mode))
+                               ll_i2info(inode)->lli_def_stripe_offset = -1;
                        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_OPEN)
-                        ll_have_md_lock(inode, &bits, mode);
-
-               fid = ll_inode2fid(inode);
-               if (!fid_res_name_eq(fid, &lock->l_resource->lr_name))
-                       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);
-                }
+                       ll_have_md_lock(inode, &bits, LCK_MINMODE);
 
-               lli = ll_i2info(inode);
                if (bits & MDS_INODELOCK_LAYOUT) {
-                       struct cl_object_conf conf = { { 0 } };
+                       struct cl_object_conf conf = {
+                               .coc_opc = OBJECT_CONF_INVALIDATE,
+                               .coc_inode = inode,
+                       };
 
-                       conf.coc_opc = OBJECT_CONF_INVALIDATE;
-                       conf.coc_inode = inode;
                        rc = ll_layout_conf(inode, &conf);
-                       if (rc)
-                               CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
+                       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;
+                       lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
                        spin_unlock(&lli->lli_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_invalidate_negative_children(inode);
+               if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
+                       struct ll_inode_info *lli = ll_i2info(inode);
+
+                       CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
+                              "pfid  = "DFID"\n", PFID(ll_inode2fid(inode)),
+                              lli, PFID(&lli->lli_pfid));
+                       truncate_inode_pages(inode->i_mapping, 0);
+
+                       if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
+                               struct inode *master_inode = NULL;
+                               unsigned long hash;
+
+                               /* This is slave inode, since all of the child
+                                * dentry is connected on the master inode, so
+                                * we have to invalidate the negative children
+                                * on master inode */
+                               CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
+                                      PFID(ll_inode2fid(inode)),
+                                      PFID(&lli->lli_pfid));
+
+                               hash = cl_fid_build_ino(&lli->lli_pfid,
+                                       ll_need_32bit_api(ll_i2sbi(inode)));
+
+                               master_inode = ilookup5(inode->i_sb, hash,
+                                                       ll_test_inode_by_fid,
+                                                       (void *)&lli->lli_pfid);
+                               if (master_inode != NULL &&
+                                       !IS_ERR(master_inode)) {
+                                       ll_invalidate_negative_children(
+                                                               master_inode);
+                                       iput(master_inode);
+                               }
+                       } else {
+                               ll_invalidate_negative_children(inode);
+                       }
                }
 
-               if (inode->i_sb->s_root &&
-                   inode != inode->i_sb->s_root->d_inode &&
-                   (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)))
+               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);
+               iput(inode);
+               break;
+       }
+       default:
+               LBUG();
+       }
+
+       RETURN(0);
 }
 
 __u32 ll_i2suppgid(struct inode *i)
 {
        if (in_group_p(i->i_gid))
-               return (__u32)i->i_gid;
+               return (__u32)from_kgid(&init_user_ns, i->i_gid);
        else
-               return (__u32)(-1);
+               return (__u32) __kgid_val(INVALID_GID);
 }
 
 /* Pack the required supplementary groups into the supplied groups array.
@@ -318,35 +338,15 @@ __u32 ll_i2suppgid(struct inode *i)
  * array in case it might be useful.  Not needed if doing an MDS-side upcall. */
 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
 {
-#if 0
-        int i;
-#endif
+       LASSERT(i1 != NULL);
+       LASSERT(suppgids != NULL);
 
-        LASSERT(i1 != NULL);
-        LASSERT(suppgids != NULL);
+       suppgids[0] = ll_i2suppgid(i1);
 
-        suppgids[0] = ll_i2suppgid(i1);
-
-        if (i2)
-                suppgids[1] = ll_i2suppgid(i2);
-                else
-                        suppgids[1] = -1;
-
-#if 0
-        for (i = 0; i < current_ngroups; i++) {
-                if (suppgids[0] == -1) {
-                        if (current_groups[i] != suppgids[1])
-                                suppgids[0] = current_groups[i];
-                        continue;
-                }
-                if (suppgids[1] == -1) {
-                        if (current_groups[i] != suppgids[0])
-                                suppgids[1] = current_groups[i];
-                        continue;
-                }
-                break;
-        }
-#endif
+       if (i2)
+               suppgids[1] = ll_i2suppgid(i2);
+       else
+               suppgids[1] = -1;
 }
 
 /*
@@ -418,7 +418,7 @@ struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
                        iput(inode);
                        CDEBUG(D_DENTRY,
                               "Reuse dentry %p inode %p refc %d flags %#x\n",
-                             new, new->d_inode, d_refcount(new), new->d_flags);
+                             new, new->d_inode, ll_d_count(new), new->d_flags);
                        return new;
                }
        }
@@ -427,16 +427,14 @@ struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
                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_refcount(de), de->d_flags);
+              de, de->d_inode, ll_d_count(de), de->d_flags);
         return de;
 }
 
-int ll_lookup_it_finish(struct ptlrpc_request *request,
-                        struct lookup_intent *it, void *data)
+static int ll_lookup_it_finish(struct ptlrpc_request *request,
+                              struct lookup_intent *it,
+                              struct inode *parent, struct dentry **de)
 {
-       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;
@@ -471,8 +469,15 @@ int ll_lookup_it_finish(struct ptlrpc_request *request,
 
                alias = ll_splice_alias(inode, *de);
                if (IS_ERR(alias))
-                       RETURN(PTR_ERR(alias));
+                       GOTO(out, rc = 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)) {
@@ -488,25 +493,41 @@ int ll_lookup_it_finish(struct ptlrpc_request *request,
                struct lookup_intent parent_it = {
                                        .it_op = IT_GETATTR,
                                        .d.lustre.it_lock_handle = 0 };
+               struct lu_fid   fid = ll_i2info(parent)->lli_fid;
+
+               /* If it is striped directory, get the real stripe parent */
+               if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
+                       rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
+                                                ll_i2info(parent)->lli_lsm_md,
+                                                (*de)->d_name.name,
+                                                (*de)->d_name.len, &fid);
+                       if (rc != 0)
+                               GOTO(out, rc);
+               }
 
-               if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
-                                      &ll_i2info(parent)->lli_fid, NULL)) {
+               if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
+                                      NULL)) {
                        d_lustre_revalidate(*de);
                        ll_intent_release(&parent_it);
                }
-        }
+       }
+
+       GOTO(out, rc = 0);
+
+out:
+       if (rc != 0 && it->it_op & IT_OPEN)
+               ll_open_cleanup((*de)->d_sb, request);
 
-        RETURN(0);
+       return rc;
 }
 
 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
-                                   struct lookup_intent *it, int lookup_flags)
+                                  struct lookup_intent *it)
 {
-        struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
-        struct dentry *save = dentry, *retval;
-        struct ptlrpc_request *req = NULL;
-        struct md_op_data *op_data;
-        struct it_cb_data icbd;
+       struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
+       struct dentry *save = dentry, *retval;
+       struct ptlrpc_request *req = NULL;
+       struct md_op_data *op_data = NULL;
         __u32 opc;
         int rc;
         ENTRY;
@@ -514,49 +535,68 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
         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);
-
-        if (it->it_op == IT_GETATTR) {
-                rc = ll_statahead_enter(parent, &dentry, 0);
-                if (rc == 1) {
-                        if (dentry == save)
-                                GOTO(out, retval = NULL);
-                        GOTO(out, retval = dentry);
-                }
-        }
+       if (it == NULL || it->it_op == IT_GETXATTR)
+               it = &lookup_it;
 
-        icbd.icbd_childp = &dentry;
-        icbd.icbd_parent = parent;
+       if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
+               rc = ll_statahead(parent, &dentry, 0);
+               if (rc == 1)
+                       RETURN(dentry == save ? NULL : dentry);
+       }
 
        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,
-                                     NULL);
-        if (IS_ERR(op_data))
-                RETURN((void *)op_data);
+       op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
+                                    dentry->d_name.len, 0, opc, NULL);
+       if (IS_ERR(op_data))
+               RETURN((void *)op_data);
 
-        /* enforce umask if acl disabled or MDS doesn't support umask */
-        if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
+       /* 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);
-        ll_finish_md_op_data(op_data);
-        if (rc < 0)
-                GOTO(out, retval = ERR_PTR(rc));
+       rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
+                           &ll_md_blocking_ast, 0);
+       /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
+        * client does not know which suppgid should be sent to the MDS, or
+        * some other(s) changed the target file's GID after this RPC sent
+        * to the MDS with the suppgid as the original GID, then we should
+        * try again with right suppgid. */
+       if (rc == -EACCES && it->it_op & IT_OPEN &&
+           it_disposition(it, DISP_OPEN_DENY)) {
+               struct mdt_body *body;
+
+               LASSERT(req != NULL);
+
+               body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+               if (op_data->op_suppgids[0] == body->mbo_gid ||
+                   op_data->op_suppgids[1] == body->mbo_gid ||
+                   !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
+                       GOTO(out, retval = ERR_PTR(-EACCES));
+
+               fid_zero(&op_data->op_fid2);
+               op_data->op_suppgids[1] = body->mbo_gid;
+               ptlrpc_req_finished(req);
+               req = NULL;
+               ll_intent_release(it);
+               rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
+                                   &ll_md_blocking_ast, 0);
+       }
+
+       if (rc < 0)
+               GOTO(out, retval = ERR_PTR(rc));
 
-        rc = ll_lookup_it_finish(req, it, &icbd);
+       rc = ll_lookup_it_finish(req, it, parent, &dentry);
         if (rc != 0) {
                 ll_intent_release(it);
                 GOTO(out, retval = ERR_PTR(rc));
@@ -569,16 +609,14 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
         }
         ll_lookup_finish_locks(it, dentry);
 
-        if (dentry == save)
-                GOTO(out, retval = NULL);
-        else
-                GOTO(out, retval = 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;
+       GOTO(out, retval = (dentry == save) ? NULL : dentry);
+
+out:
+       if (op_data != NULL && !IS_ERR(op_data))
+               ll_finish_md_op_data(op_data);
+
+       ptlrpc_req_finished(req);
+       return retval;
 }
 
 #ifdef HAVE_IOP_ATOMIC_OPEN
@@ -588,9 +626,9 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
        struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
        struct dentry *de;
 
-       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),flags=%u\n",
-              dentry->d_name.len, dentry->d_name.name, parent->i_ino,
-              parent->i_generation, parent, flags);
+       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);
 
        /* Optimize away (CREATE && !OPEN). Let .create handle the race. */
        if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN))
@@ -600,7 +638,7 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
                itp = NULL;
        else
                itp = &it;
-       de = ll_lookup_it(parent, dentry, itp, 0);
+       de = ll_lookup_it(parent, dentry, itp);
 
        if (itp != NULL)
                ll_intent_release(itp);
@@ -622,10 +660,10 @@ static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
        int rc = 0;
        ENTRY;
 
-       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),file %p,"
-                          "open_flags %x,mode %x opened %d\n",
-              dentry->d_name.len, dentry->d_name.name, dir->i_ino,
-              dir->i_generation, dir, file, open_flags, mode, *opened);
+       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)
@@ -638,9 +676,10 @@ static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
        }
        it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
        it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
+       it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 
        /* Dentry added to dcache tree in ll_lookup_it */
-       de = ll_lookup_it(dir, dentry, it, lookup_flags);
+       de = ll_lookup_it(dir, dentry, it);
        if (IS_ERR(de))
                rc = PTR_ERR(de);
        else if (de != NULL)
@@ -649,7 +688,7 @@ static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
        if (!rc) {
                if (it_disposition(it, DISP_OPEN_CREATE)) {
                        /* Dentry instantiated in ll_create_it. */
-                       rc = ll_create_it(dir, dentry, mode, it);
+                       rc = ll_create_it(dir, dentry, it);
                        if (rc) {
                                /* We dget in ll_splice_alias. */
                                if (de != NULL)
@@ -692,12 +731,12 @@ out_release:
 }
 
 #else /* !HAVE_IOP_ATOMIC_OPEN */
-struct lookup_intent *ll_convert_intent(struct open_intent *oit,
-                                        int lookup_flags)
+static struct lookup_intent *
+ll_convert_intent(struct open_intent *oit, int lookup_flags)
 {
        struct lookup_intent *it;
 
-       OBD_ALLOC(it, sizeof(*it));
+       OBD_ALLOC_PTR(it);
        if (!it)
                return ERR_PTR(-ENOMEM);
 
@@ -707,6 +746,7 @@ struct lookup_intent *ll_convert_intent(struct open_intent *oit,
                        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);
+               it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
        } else {
                it->it_op = IT_GETATTR;
        }
@@ -736,7 +776,7 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
                                 RETURN((struct dentry *)it);
                 }
 
-                de = ll_lookup_it(parent, dentry, it, nd->flags);
+               de = ll_lookup_it(parent, dentry, it);
                 if (de)
                         dentry = de;
                 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
@@ -777,18 +817,15 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
                         OBD_FREE(it, sizeof(*it));
                 }
         } else {
-                de = ll_lookup_it(parent, dentry, NULL, 0);
-        }
+               de = ll_lookup_it(parent, dentry, NULL);
+       }
 
-        RETURN(de);
+       RETURN(de);
 }
 #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,
-                                    int namelen, const void *data, int datalen,
-                                    int mode, __u64 extra,
-                                    struct lookup_intent *it)
+static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 {
         struct inode *inode = NULL;
         struct ptlrpc_request *request = NULL;
@@ -810,8 +847,8 @@ static struct inode *ll_create_node(struct inode *dir, const char *name,
         /* 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);
+       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:
@@ -833,55 +870,52 @@ static struct inode *ll_create_node(struct inode *dir, const char *name,
  * If the create succeeds, we fill in the inode information
  * with d_instantiate().
  */
-static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
+static int ll_create_it(struct inode *dir, struct dentry *dentry,
                        struct lookup_intent *it)
 {
        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);
 
-       inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
-                              NULL, 0, mode, 0, it);
+       inode = ll_create_node(dir, it);
        if (IS_ERR(inode))
                RETURN(PTR_ERR(inode));
 
-       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,
-                            struct inode *inode)
+void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
 {
-        struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
-                                                       &RMF_MDT_BODY);
-
-        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);
-                LTIME_S(inode->i_mtime) = body->mtime;
-        }
-        if (body->valid & OBD_MD_FLCTIME &&
-            body->ctime > LTIME_S(inode->i_ctime))
-                LTIME_S(inode->i_ctime) = body->ctime;
+       struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
+                                                      &RMF_MDT_BODY);
+
+       LASSERT(body);
+       if (body->mbo_valid & OBD_MD_FLMTIME &&
+           body->mbo_mtime > LTIME_S(inode->i_mtime)) {
+               CDEBUG(D_INODE, "setting fid "DFID" mtime from %lu to "LPU64
+                      "\n", PFID(ll_inode2fid(inode)),
+                      LTIME_S(inode->i_mtime), body->mbo_mtime);
+               LTIME_S(inode->i_mtime) = body->mbo_mtime;
+       }
+
+       if (body->mbo_valid & OBD_MD_FLCTIME &&
+           body->mbo_ctime > LTIME_S(inode->i_ctime))
+               LTIME_S(inode->i_ctime) = body->mbo_ctime;
 }
 
-static int ll_new_node(struct inode *dir, struct qstr *name,
-                       const char *tgt, int mode, int rdev,
-                       struct dentry *dchild, __u32 opc)
+static int ll_new_node(struct inode *dir, struct dentry *dchild,
+                      const char *tgt, umode_t mode, int rdev, __u32 opc)
 {
+       struct qstr *name = &dchild->d_name;
         struct ptlrpc_request *request = NULL;
         struct md_op_data *op_data;
         struct inode *inode = NULL;
@@ -893,42 +927,78 @@ static int ll_new_node(struct inode *dir, struct qstr *name,
         if (unlikely(tgt != NULL))
                 tgt_len = strlen(tgt) + 1;
 
+again:
         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
                                      name->len, 0, opc, NULL);
         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,
-                       current_fsuid(), current_fsgid(),
+                       from_kuid(&init_user_ns, current_fsuid()),
+                       from_kgid(&init_user_ns, current_fsgid()),
                        cfs_curproc_cap_pack(), rdev, &request);
        ll_finish_md_op_data(op_data);
-        if (err)
-                GOTO(err_exit, err);
+       if (err) {
+               /* If the client doesn't know where to create a subdirectory (or
+                * in case of a race that sends the RPC to the wrong MDS), the
+                * MDS will return -EREMOTE and the client will fetch the layout
+                * for the directory, either from the local xattr cache or the
+                * MDS, then create the directory on the right MDT. */
+               if (err == -EREMOTE) {
+                       struct lmv_user_md *lum;
+                       int rc;
+
+                       ptlrpc_req_finished(request);
+                       request = NULL;
+
+                       OBD_ALLOC_PTR(lum);
+                       if (lum == NULL)
+                               GOTO(err_exit, err = -ENOMEM);
+
+                       rc = ll_getxattr_common(dir, XATTR_NAME_DEFAULT_LMV,
+                                               lum, sizeof(*lum),
+                                               OBD_MD_FLXATTR);
+                       if (rc < 0) {
+                               OBD_FREE_PTR(lum);
+                               if (rc == -ENODATA)
+                                       GOTO(err_exit, err);
+                               else
+                                       GOTO(err_exit, rc);
+                       }
+
+                       ll_i2info(dir)->lli_def_stripe_offset =
+                               le32_to_cpu(lum->lum_stripe_offset);
+                       OBD_FREE_PTR(lum);
+                       goto again;
+               }
+               GOTO(err_exit, err);
+       }
 
         ll_update_times(request, dir);
 
-        if (dchild) {
-                err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
-                if (err)
-                     GOTO(err_exit, err);
+       err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
+       if (err)
+               GOTO(err_exit, err);
+
+       d_instantiate(dchild, inode);
 
-                d_instantiate(dchild, inode);
-        }
         EXIT;
 err_exit:
-        ptlrpc_req_finished(request);
+       if (request != NULL)
+               ptlrpc_req_finished(request);
 
         return err;
 }
 
-static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
-                            unsigned rdev, struct dentry *dchild)
+static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
+                   dev_t rdev)
 {
-        int err;
+       struct qstr *name = &dchild->d_name;
+       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);
 
        if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
@@ -942,8 +1012,8 @@ static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
         case S_IFBLK:
         case S_IFIFO:
         case S_IFSOCK:
-                err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
-                                  LUSTRE_OPC_MKNOD);
+               err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
+                                 LUSTRE_OPC_MKNOD);
                 break;
         case S_IFDIR:
                 err = -EPERM;
@@ -967,12 +1037,12 @@ static int ll_create_nd(struct inode *dir, struct dentry *dentry,
 {
        int rc;
 
-       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),"
-                          "flags=%u, excl=%d\n",
-              dentry->d_name.len, dentry->d_name.name, dir->i_ino,
-              dir->i_generation, dir, mode, want_excl);
+       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);
+       rc = ll_mknod(dir, dentry, mode, 0);
 
        ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
 
@@ -993,7 +1063,7 @@ static int ll_create_nd(struct inode *dir, struct dentry *dentry,
                it = lld->lld_it;
 
         if (!it)
-                return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
+               return ll_mknod(dir, dentry, mode, 0);
 
        lld->lld_it = NULL;
 
@@ -1003,7 +1073,7 @@ static int ll_create_nd(struct inode *dir, struct dentry *dentry,
                 goto out;
         }
 
-        rc = ll_create_it(dir, dentry, mode, it);
+       rc = ll_create_it(dir, dentry, it);
         if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
                struct file *filp;
 
@@ -1024,18 +1094,19 @@ out:
 }
 #endif /* HAVE_IOP_ATOMIC_OPEN */
 
-static int ll_symlink_generic(struct inode *dir, struct qstr *name,
-                              const char *tgt, struct dentry *dchild)
+static int ll_symlink(struct inode *dir, struct dentry *dchild,
+                     const char *oldpath)
 {
-        int err;
-        ENTRY;
+       struct qstr *name = &dchild->d_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, oldpath);
 
-        err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
-                          0, dchild, LUSTRE_OPC_SYMLINK);
+       err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
+                         LUSTRE_OPC_SYMLINK);
 
         if (!err)
                 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
@@ -1043,19 +1114,20 @@ static int ll_symlink_generic(struct inode *dir, struct qstr *name,
         RETURN(err);
 }
 
-static int ll_link_generic(struct inode *src,  struct inode *dir,
-                           struct qstr *name, struct dentry *dchild)
+static int ll_link(struct dentry *old_dentry, struct inode *dir,
+                  struct dentry *new_dentry)
 {
-        struct ll_sb_info *sbi = ll_i2sbi(dir);
-        struct ptlrpc_request *request = NULL;
-        struct md_op_data *op_data;
-        int err;
+       struct inode *src = old_dentry->d_inode;
+       struct qstr *name = &new_dentry->d_name;
+       struct ll_sb_info *sbi = ll_i2sbi(dir);
+       struct ptlrpc_request *request = NULL;
+       struct md_op_data *op_data;
+       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);
+       ENTRY;
+       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);
@@ -1075,39 +1147,39 @@ out:
         RETURN(err);
 }
 
-static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
-                            int mode, struct dentry *dchild)
-
+static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
 {
+       struct qstr *name = &dchild->d_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);
 
        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);
 
-        if (!err)
-                ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
+       err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
+       if (err == 0)
+               ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
 
-        RETURN(err);
+       RETURN(err);
 }
 
-static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
-                            struct dentry *dchild, struct qstr *name)
+static int ll_rmdir(struct inode *dir, struct dentry *dchild)
 {
+       struct qstr *name = &dchild->d_name;
         struct ptlrpc_request *request = NULL;
         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);
 
-        if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
+       if (unlikely(d_mountpoint(dchild)))
                 RETURN(-EBUSY);
 
         op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
@@ -1115,8 +1187,9 @@ static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
         if (IS_ERR(op_data))
                 RETURN(PTR_ERR(op_data));
 
-       if (dchild != NULL && dchild->d_inode != NULL)
+       if (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);
@@ -1139,8 +1212,8 @@ int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
        int rc;
        ENTRY;
 
-       CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
-              namelen, name, dir->i_ino, dir->i_generation, dir);
+       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);
@@ -1158,109 +1231,33 @@ int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
        RETURN(rc);
 }
 
-int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
-{
-        struct mdt_body *body;
-        struct lov_mds_md *eadata;
-        struct lov_stripe_md *lsm = NULL;
-        struct obd_trans_info oti = { 0 };
-        struct obdo *oa;
-        struct obd_capa *oc = NULL;
-        int rc;
-        ENTRY;
-
-        /* req is swabbed so this is safe */
-        body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
-        if (!(body->valid & OBD_MD_FLEASIZE))
-                RETURN(0);
-
-        if (body->eadatasize == 0) {
-                CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
-                GOTO(out, rc = -EPROTO);
-        }
-
-        /* The MDS sent back the EA because we unlinked the last reference
-         * to this file. Use this EA to unlink the objects on the OST.
-         * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
-         * check it is complete and sensible. */
-        eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
-                                              body->eadatasize);
-        LASSERT(eadata != NULL);
-
-        rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
-        if (rc < 0) {
-                CERROR("obd_unpackmd: %d\n", rc);
-                GOTO(out, rc);
-        }
-        LASSERT(rc >= sizeof(*lsm));
-
-        OBDO_ALLOC(oa);
-        if (oa == NULL)
-                GOTO(out_free_memmd, rc = -ENOMEM);
-
-       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;
-
-        if (body->valid & OBD_MD_FLCOOKIE) {
-                oa->o_valid |= OBD_MD_FLCOOKIE;
-                oti.oti_logcookies =
-                        req_capsule_server_sized_get(&request->rq_pill,
-                                                     &RMF_LOGCOOKIES,
-                                                   sizeof(struct llog_cookie) *
-                                                     lsm->lsm_stripe_count);
-                if (oti.oti_logcookies == NULL) {
-                        oa->o_valid &= ~OBD_MD_FLCOOKIE;
-                        body->valid &= ~OBD_MD_FLCOOKIE;
-                }
-        }
-
-        if (body->valid & OBD_MD_FLOSSCAPA) {
-                rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
-                if (rc)
-                        GOTO(out_free_memmd, 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.
- * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
- * is any lock existing. They will recycle dentries and inodes based upon locks
- * too. b=20433 */
-static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
-                             struct dentry *dchild, struct qstr *name)
+/* ll_unlink() doesn't update the inode with the new link count.
+ * Instead, ll_ddelete() and ll_d_iput() will update it based upon if
+ * there is any lock existing. They will recycle dentries and inodes
+ * based upon locks too. b=20433 */
+static int ll_unlink(struct inode *dir, struct dentry *dchild)
 {
+       struct qstr *name = &dchild->d_name;
         struct ptlrpc_request *request = NULL;
         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,
          * just check it as vfs_unlink does.
          */
-        if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
-                RETURN(-EBUSY);
+       if (unlikely(d_mountpoint(dchild)))
+               RETURN(-EBUSY);
 
-        op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
-                                     name->len, 0, LUSTRE_OPC_ANY, NULL);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
+       op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
+                                    LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
 
-       if (dchild != NULL && dchild->d_inode != NULL)
+       if (dchild->d_inode != NULL)
                op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
 
        op_data->op_fid2 = op_data->op_fid3;
@@ -1272,39 +1269,39 @@ static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
         ll_update_times(request, dir);
         ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
 
-        rc = ll_objects_destroy(request, dir);
  out:
         ptlrpc_req_finished(request);
         RETURN(rc);
 }
 
-static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
-                             struct dentry *src_dchild, struct qstr *src_name,
-                             struct inode *tgt, struct dentry *tgt_dparent,
-                             struct dentry *tgt_dchild, struct qstr *tgt_name)
+static int ll_rename(struct inode *src, struct dentry *src_dchild,
+                    struct inode *tgt, struct dentry *tgt_dchild)
 {
+       struct qstr *src_name = &src_dchild->d_name;
+       struct qstr *tgt_name = &tgt_dchild->d_name;
         struct ptlrpc_request *request = NULL;
         struct ll_sb_info *sbi = ll_i2sbi(src);
         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)))
-                RETURN(-EBUSY);
+       if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
+               RETURN(-EBUSY);
 
-        op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
-                                     LUSTRE_OPC_ANY, NULL);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
+       op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
+                                    LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
 
-       if (src_dchild != NULL && src_dchild->d_inode != NULL)
+       if (src_dchild->d_inode != NULL)
                op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
-       if (tgt_dchild != NULL && tgt_dchild->d_inode != NULL)
+
+       if (tgt_dchild->d_inode != NULL)
                op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
 
         err = md_rename(sbi->ll_md_exp, op_data,
@@ -1315,63 +1312,17 @@ static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
                 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);
         }
 
         ptlrpc_req_finished(request);
 
-        RETURN(err);
-}
+       if (err == 0)
+               d_move(src_dchild, tgt_dchild);
 
-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);
-}
-
-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, 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)
-{
-        int err;
-        err = ll_rename_generic(old_dir, NULL,
-                                 old_dentry, &old_dentry->d_name,
-                                 new_dir, NULL, new_dentry,
-                                 &new_dentry->d_name);
-       if (!err)
-                        d_move(old_dentry, new_dentry);
-        return err;
+       RETURN(err);
 }
 
-struct inode_operations ll_dir_inode_operations = {
+const struct inode_operations ll_dir_inode_operations = {
        .mknod              = ll_mknod,
 #ifdef HAVE_IOP_ATOMIC_OPEN
        .atomic_open        = ll_atomic_open,
@@ -1397,7 +1348,7 @@ struct inode_operations ll_dir_inode_operations = {
 #endif
 };
 
-struct inode_operations ll_special_inode_operations = {
+const struct inode_operations ll_special_inode_operations = {
        .setattr        = ll_setattr,
        .getattr        = ll_getattr,
        .permission     = ll_inode_permission,