Whamcloud - gitweb
LU-333: Make read_bytes in llite/*/stats report bytes read.
[fs/lustre-release.git] / lustre / llite / file.c
index beaf065..93d4f47 100644 (file)
@@ -26,7 +26,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  */
 /*
@@ -75,12 +75,18 @@ void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
         op_data->op_attr.ia_ctime = inode->i_ctime;
         op_data->op_attr.ia_size = i_size_read(inode);
         op_data->op_attr_blocks = inode->i_blocks;
-        ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = inode->i_flags;
+        ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags =
+                                        ll_inode_to_ext_flags(inode->i_flags);
         op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
-        memcpy(&op_data->op_handle, fh, sizeof(op_data->op_handle));
+        if (fh)
+                op_data->op_handle = *fh;
         op_data->op_capa1 = ll_mdscapa_get(inode);
 }
 
+/**
+ * Closes the IO epoch and packs all the attributes into @op_data for
+ * the CLOSE rpc.
+ */
 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
                              struct obd_client_handle *och)
 {
@@ -92,14 +98,15 @@ static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
         if (!(och->och_flags & FMODE_WRITE))
                 goto out;
 
-        if (!(ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) ||
-            !S_ISREG(inode->i_mode))
+        if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode))
                 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
         else
-                ll_epoch_close(inode, op_data, &och, 0);
+                ll_ioepoch_close(inode, op_data, &och, 0);
 
 out:
         ll_pack_inode2opdata(inode, op_data, &och->och_fh);
+        ll_prep_md_op_data(op_data, inode, NULL, NULL,
+                           0, 0, LUSTRE_OPC_ANY, NULL);
         EXIT;
 }
 
@@ -125,14 +132,6 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp,
                 GOTO(out, rc = 0);
         }
 
-        /*
-         * here we check if this is forced umount. If so this is called on
-         * canceling "open lock" and we do not call md_close() in this case, as
-         * it will not be successful, as import is already deactivated.
-         */
-        if (obd->obd_force)
-                GOTO(out, rc = 0);
-
         OBD_ALLOC_PTR(op_data);
         if (op_data == NULL)
                 GOTO(out, rc = -ENOMEM); // XXX We leak openhandle and request here.
@@ -142,12 +141,10 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp,
         rc = md_close(md_exp, op_data, och->och_mod, &req);
         if (rc == -EAGAIN) {
                 /* This close must have the epoch closed. */
-                LASSERT(exp->exp_connect_flags & OBD_CONNECT_SOM);
                 LASSERT(epoch_close);
                 /* MDS has instructed us to obtain Size-on-MDS attribute from
                  * OSTs and send setattr to back to MDS. */
-                rc = ll_sizeonmds_update(inode, &och->och_fh,
-                                         op_data->op_ioepoch);
+                rc = ll_som_update(inode, op_data);
                 if (rc) {
                         CERROR("inode %lu mdc Size-on-MDS update failed: "
                                "rc = %d\n", inode->i_ino, rc);
@@ -169,7 +166,7 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp,
         EXIT;
 out:
 
-        if ((exp->exp_connect_flags & OBD_CONNECT_SOM) && !epoch_close &&
+        if (exp_connect_som(exp) && !epoch_close &&
             S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
                 ll_queue_done_writing(inode, LLIF_DONE_WRITING);
         } else {
@@ -204,15 +201,15 @@ int ll_md_real_close(struct inode *inode, int flags)
                 och_usecount = &lli->lli_open_fd_read_count;
         }
 
-        down(&lli->lli_och_sem);
+        cfs_down(&lli->lli_och_sem);
         if (*och_usecount) { /* There are still users of this handle, so
                                 skip freeing it. */
-                up(&lli->lli_och_sem);
+                cfs_up(&lli->lli_och_sem);
                 RETURN(0);
         }
         och=*och_p;
         *och_p = NULL;
-        up(&lli->lli_och_sem);
+        cfs_up(&lli->lli_och_sem);
 
         if (och) { /* There might be a race and somebody have freed this och
                       already */
@@ -244,7 +241,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode,
                 struct inode *inode = file->f_dentry->d_inode;
                 ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}};
 
-                down(&lli->lli_och_sem);
+                cfs_down(&lli->lli_och_sem);
                 if (fd->fd_omode & FMODE_WRITE) {
                         lockmode = LCK_CW;
                         LASSERT(lli->lli_open_fd_write_count);
@@ -258,7 +255,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode,
                         LASSERT(lli->lli_open_fd_read_count);
                         lli->lli_open_fd_read_count--;
                 }
-                up(&lli->lli_och_sem);
+                cfs_up(&lli->lli_och_sem);
 
                 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
                                    LDLM_IBITS, &policy, lockmode,
@@ -333,6 +330,10 @@ int ll_file_release(struct inode *inode, struct file *file)
         lli->lli_async_rc = 0;
 
         rc = ll_md_close(sbi->ll_md_exp, inode, file);
+
+        if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
+                libcfs_debug_dumplog();
+
         RETURN(rc);
 }
 
@@ -345,6 +346,7 @@ static int ll_intent_file_open(struct file *file, void *lmm,
         const int len = file->f_dentry->d_name.len;
         struct md_op_data *op_data;
         struct ptlrpc_request *req;
+        __u32 opc = LUSTRE_OPC_ANY;
         int rc;
         ENTRY;
 
@@ -360,12 +362,15 @@ static int ll_intent_file_open(struct file *file, void *lmm,
          * makes a good candidate for using OPEN lock */
         /* If lmmsize & lmm are not 0, we are just setting stripe info
          * parameters. No need for the open lock */
-        if (!lmm && !lmmsize)
+        if (lmm == NULL && lmmsize == 0) {
                 itp->it_flags |= MDS_OPEN_LOCK;
+                if (itp->it_flags & FMODE_WRITE)
+                        opc = LUSTRE_OPC_CREATE;
+        }
 
         op_data  = ll_prep_md_op_data(NULL, parent->d_inode,
                                       file->f_dentry->d_inode, name, len,
-                                      O_RDWR, LUSTRE_OPC_ANY, NULL);
+                                      O_RDWR, opc, NULL);
         if (IS_ERR(op_data))
                 RETURN(PTR_ERR(op_data));
 
@@ -389,12 +394,12 @@ static int ll_intent_file_open(struct file *file, void *lmm,
                 GOTO(out, rc);
         }
 
-        if (itp->d.lustre.it_lock_mode)
+        rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL);
+        if (!rc && itp->d.lustre.it_lock_mode)
                 md_set_lock_data(sbi->ll_md_exp,
                                  &itp->d.lustre.it_lock_handle,
-                                 file->f_dentry->d_inode);
+                                 file->f_dentry->d_inode, NULL);
 
-        rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL);
 out:
         ptlrpc_req_finished(itp->d.lustre.it_data);
         it_clear_disposition(itp, DISP_ENQ_COMPLETE);
@@ -403,6 +408,20 @@ out:
         RETURN(rc);
 }
 
+/**
+ * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
+ * not believe attributes if a few ioepoch holders exist. Attributes for
+ * previous ioepoch if new one is opened are also skipped by MDS.
+ */
+void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
+{
+        if (ioepoch && lli->lli_ioepoch != ioepoch) {
+                lli->lli_ioepoch = ioepoch;
+                CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n",
+                       ioepoch, PFID(&lli->lli_fid));
+        }
+}
+
 static int ll_och_fill(struct obd_export *md_exp, struct ll_inode_info *lli,
                        struct lookup_intent *it, struct obd_client_handle *och)
 {
@@ -418,7 +437,7 @@ static int ll_och_fill(struct obd_export *md_exp, struct ll_inode_info *lli,
         och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
         och->och_fid = lli->lli_fid;
         och->och_flags = it->it_flags;
-        lli->lli_ioepoch = body->ioepoch;
+        ll_ioepoch_open(lli, body->ioepoch);
 
         return md_set_open_replay_data(md_exp, och, req);
 }
@@ -487,12 +506,8 @@ int ll_file_open(struct inode *inode, struct file *file)
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), flags %o\n", inode->i_ino,
                inode->i_generation, inode, file->f_flags);
 
-#ifdef HAVE_VFS_INTENT_PATCHES
-        it = file->f_it;
-#else
         it = file->private_data; /* XXX: compat macro */
         file->private_data = NULL; /* prevent ll_local_open assertion */
-#endif
 
         fd = ll_file_data_get();
         if (fd == NULL)
@@ -500,31 +515,14 @@ int ll_file_open(struct inode *inode, struct file *file)
 
         fd->fd_file = file;
         if (S_ISDIR(inode->i_mode)) {
-again:
-                spin_lock(&lli->lli_lock);
+                cfs_spin_lock(&lli->lli_sa_lock);
                 if (lli->lli_opendir_key == NULL && lli->lli_opendir_pid == 0) {
                         LASSERT(lli->lli_sai == NULL);
                         lli->lli_opendir_key = fd;
                         lli->lli_opendir_pid = cfs_curproc_pid();
                         opendir_set = 1;
-                } else if (unlikely(lli->lli_opendir_pid == cfs_curproc_pid() &&
-                                    lli->lli_opendir_key != NULL)) {
-                        /* Two cases for this:
-                         * (1) The same process open such directory many times.
-                         * (2) The old process opened the directory, and exited
-                         *     before its children processes. Then new process
-                         *     with the same pid opens such directory before the
-                         *     old process's children processes exit.
-                         * reset stat ahead for such cases. */
-                        spin_unlock(&lli->lli_lock);
-                        CDEBUG(D_INFO, "Conflict statahead for %.*s "DFID
-                               " reset it.\n", file->f_dentry->d_name.len,
-                               file->f_dentry->d_name.name,
-                               PFID(&lli->lli_fid));
-                        ll_stop_statahead(inode, lli->lli_opendir_key);
-                        goto again;
                 }
-                spin_unlock(&lli->lli_lock);
+                cfs_spin_unlock(&lli->lli_sa_lock);
         }
 
         if (inode->i_sb->s_root == file->f_dentry) {
@@ -545,13 +543,19 @@ again:
                  * dentry_open after call to open_namei that checks permissions.
                  * Only nfsd_open call dentry_open directly without checking
                  * permissions and because of that this code below is safe. */
-                if (oit.it_flags & FMODE_WRITE)
+                if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
                         oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
 
                 /* We do not want O_EXCL here, presumably we opened the file
                  * already? XXX - NFS implications? */
                 oit.it_flags &= ~O_EXCL;
 
+                /* bug20584, if "it_flags" contains O_CREAT, the file will be
+                 * created if necessary, then "IT_CREAT" should be set to keep
+                 * consistent with it */
+                if (oit.it_flags & O_CREAT)
+                        oit.it_op |= IT_CREAT;
+
                 it = &oit;
         }
 
@@ -568,14 +572,14 @@ restart:
                 och_usecount = &lli->lli_open_fd_read_count;
         }
 
-        down(&lli->lli_och_sem);
+        cfs_down(&lli->lli_och_sem);
         if (*och_p) { /* Open handle is present */
                 if (it_disposition(it, DISP_OPEN_OPEN)) {
                         /* Well, there's extra open request that we do not need,
                            let's close it somehow. This will decref request. */
                         rc = it_open_error(DISP_OPEN_OPEN, it);
                         if (rc) {
-                                up(&lli->lli_och_sem);
+                                cfs_up(&lli->lli_och_sem);
                                 ll_file_data_put(fd);
                                 GOTO(out_openerr, rc);
                         }
@@ -588,7 +592,7 @@ restart:
                 rc = ll_local_open(file, it, fd, NULL);
                 if (rc) {
                         (*och_usecount)--;
-                        up(&lli->lli_och_sem);
+                        cfs_up(&lli->lli_och_sem);
                         ll_file_data_put(fd);
                         GOTO(out_openerr, rc);
                 }
@@ -600,7 +604,7 @@ restart:
                            could be cancelled, and since blocking ast handler
                            would attempt to grab och_sem as well, that would
                            result in a deadlock */
-                        up(&lli->lli_och_sem);
+                        cfs_up(&lli->lli_och_sem);
                         it->it_create_mode |= M_CHECK_STALE;
                         rc = ll_intent_file_open(file, NULL, 0, it);
                         it->it_create_mode &= ~M_CHECK_STALE;
@@ -614,9 +618,6 @@ restart:
                                 req = it->d.lustre.it_data;
                                 ptlrpc_req_finished(req);
                         }
-                        md_set_lock_data(ll_i2sbi(inode)->ll_md_exp,
-                                         &it->d.lustre.it_lock_handle,
-                                         file->f_dentry->d_inode);
                         goto restart;
                 }
                 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
@@ -645,7 +646,7 @@ restart:
                         GOTO(out_och_free, rc);
                 }
         }
-        up(&lli->lli_och_sem);
+        cfs_up(&lli->lli_och_sem);
 
         /* Must do this outside lli_och_sem lock to prevent deadlock where
            different kind of OPEN lock for this same inode gets cancelled
@@ -676,7 +677,7 @@ out_och_free:
                         *och_p = NULL; /* OBD_FREE writes some magic there */
                         (*och_usecount)--;
                 }
-                up(&lli->lli_och_sem);
+                cfs_up(&lli->lli_och_sem);
 out_openerr:
                 if (opendir_set != 0)
                         ll_stop_statahead(inode, lli->lli_opendir_key);
@@ -687,7 +688,8 @@ out_openerr:
 
 /* Fills the obdo with the attributes for the lsm */
 static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
-                          struct obd_capa *capa, struct obdo *obdo)
+                          struct obd_capa *capa, struct obdo *obdo,
+                          __u64 ioepoch, int sync)
 {
         struct ptlrpc_request_set *set;
         struct obd_info            oinfo = { { { 0 } } };
@@ -700,14 +702,19 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
         oinfo.oi_md = lsm;
         oinfo.oi_oa = obdo;
         oinfo.oi_oa->o_id = lsm->lsm_object_id;
-        oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
+        oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
         oinfo.oi_oa->o_mode = S_IFREG;
+        oinfo.oi_oa->o_ioepoch = ioepoch;
         oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
                                OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
                                OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
-                               OBD_MD_FLGROUP;
+                               OBD_MD_FLGROUP | OBD_MD_FLEPOCH;
         oinfo.oi_capa = capa;
+        if (sync) {
+                oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
+                oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
+        }
 
         set = ptlrpc_prep_set();
         if (set == NULL) {
@@ -726,15 +733,20 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
         RETURN(rc);
 }
 
-/* Fills the obdo with the attributes for the inode defined by lsm */
-int ll_inode_getattr(struct inode *inode, struct obdo *obdo)
+/**
+  * Performs the getattr on the inode and updates its fields.
+  * If @sync != 0, perform the getattr under the server-side lock.
+  */
+int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
+                     __u64 ioepoch, int sync)
 {
         struct ll_inode_info *lli  = ll_i2info(inode);
         struct obd_capa      *capa = ll_mdscapa_get(inode);
         int rc;
         ENTRY;
 
-        rc = ll_lsm_getattr(lli->lli_smd, ll_i2dtexp(inode), capa, obdo);
+        rc = ll_lsm_getattr(lli->lli_smd, ll_i2dtexp(inode),
+                            capa, obdo, ioepoch, sync);
         capa_put(capa);
         if (rc == 0) {
                 obdo_refresh_inode(inode, obdo, obdo->o_valid);
@@ -758,8 +770,17 @@ int ll_merge_lvb(struct inode *inode)
 
         ll_inode_size_lock(inode, 1);
         inode_init_lvb(inode, &lvb);
+
+        /* merge timestamps the most resently obtained from mds with
+           timestamps obtained from osts */
+        lvb.lvb_atime = lli->lli_lvb.lvb_atime;
+        lvb.lvb_mtime = lli->lli_lvb.lvb_mtime;
+        lvb.lvb_ctime = lli->lli_lvb.lvb_ctime;
         rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0);
-        i_size_write(inode, lvb.lvb_size);
+        cl_isize_write_nolock(inode, lvb.lvb_size);
+
+        CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n",
+               PFID(&lli->lli_fid), lvb.lvb_size);
         inode->i_blocks = lvb.lvb_blocks;
 
         LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
@@ -776,7 +797,7 @@ int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
         struct obdo obdo = { 0 };
         int rc;
 
-        rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo);
+        rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo, 0, 0);
         if (rc == 0) {
                 st->st_size   = obdo.o_size;
                 st->st_blocks = obdo.o_blocks;
@@ -789,19 +810,14 @@ int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
 
 void ll_io_init(struct cl_io *io, const struct file *file, int write)
 {
-        struct inode *inode     = file->f_dentry->d_inode;
-        struct ll_sb_info *sbi  = ll_i2sbi(inode);
-        struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
+        struct inode *inode = file->f_dentry->d_inode;
 
-        LASSERT(fd != NULL);
-        memset(io, 0, sizeof *io);
         io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
         if (write)
-                io->u.ci_wr.wr_append = file->f_flags & O_APPEND;
+                io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
         io->ci_obj     = ll_i2info(inode)->lli_clob;
         io->ci_lockreq = CILR_MAYBE;
-        if (fd->fd_flags & LL_FILE_IGNORE_LOCK ||
-            sbi->ll_flags & LL_SBI_NOLCK) {
+        if (ll_file_nolock(file)) {
                 io->ci_lockreq = CILR_NEVER;
                 io->ci_no_srvlock = 1;
         } else if (file->f_flags & O_APPEND) {
@@ -810,43 +826,87 @@ void ll_io_init(struct cl_io *io, const struct file *file, int write)
 }
 
 static ssize_t ll_file_io_generic(const struct lu_env *env,
-                struct ccc_io_args *args, struct file *file,
+                struct vvp_io_args *args, struct file *file,
                 enum cl_io_type iot, loff_t *ppos, size_t count)
 {
-        struct cl_io       *io;
-        ssize_t             result;
+        struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode);
+        struct cl_io         *io;
+        ssize_t               result;
         ENTRY;
 
-        io = &ccc_env_info(env)->cti_io;
+        io = ccc_env_thread_io(env);
         ll_io_init(io, file, iot == CIT_WRITE);
 
-        if (iot == CIT_READ)
-                io->u.ci_rd.rd_is_sendfile = args->cia_is_sendfile;
-
         if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
                 struct vvp_io *vio = vvp_env_io(env);
                 struct ccc_io *cio = ccc_env_io(env);
-                if (cl_io_is_sendfile(io)) {
-                        vio->u.read.cui_actor = args->cia_actor;
-                        vio->u.read.cui_target = args->cia_target;
-                } else {
-                        cio->cui_iov = args->cia_iov;
-                        cio->cui_nrsegs = args->cia_nrsegs;
+                int write_sem_locked = 0;
+
+                cio->cui_fd  = LUSTRE_FPRIVATE(file);
+                vio->cui_io_subtype = args->via_io_subtype;
+
+                switch (vio->cui_io_subtype) {
+                case IO_NORMAL:
+                        cio->cui_iov = args->u.normal.via_iov;
+                        cio->cui_nrsegs = args->u.normal.via_nrsegs;
+                        cio->cui_tot_nrsegs = cio->cui_nrsegs;
 #ifndef HAVE_FILE_WRITEV
-                        cio->cui_iocb = args->cia_iocb;
+                        cio->cui_iocb = args->u.normal.via_iocb;
 #endif
+                        if ((iot == CIT_WRITE) &&
+                            !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
+                                if(cfs_down_interruptible(&lli->lli_write_sem))
+                                        GOTO(out, result = -ERESTARTSYS);
+                                write_sem_locked = 1;
+                        } else if (iot == CIT_READ) {
+                                cfs_down_read(&lli->lli_trunc_sem);
+                        }
+                        break;
+                case IO_SENDFILE:
+                        vio->u.sendfile.cui_actor = args->u.sendfile.via_actor;
+                        vio->u.sendfile.cui_target = args->u.sendfile.via_target;
+                        break;
+                case IO_SPLICE:
+                        vio->u.splice.cui_pipe = args->u.splice.via_pipe;
+                        vio->u.splice.cui_flags = args->u.splice.via_flags;
+                        break;
+                default:
+                        CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
+                        LBUG();
                 }
-                cio->cui_fd  = LUSTRE_FPRIVATE(file);
                 result = cl_io_loop(env, io);
-        } else
+                if (write_sem_locked)
+                        cfs_up(&lli->lli_write_sem);
+                else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ)
+                        cfs_up_read(&lli->lli_trunc_sem);
+        } else {
                 /* cl_io_rw_init() handled IO */
                 result = io->ci_result;
+        }
+
         if (io->ci_nob > 0) {
                 result = io->ci_nob;
                 *ppos = io->u.ci_wr.wr.crw_pos;
         }
+        GOTO(out, result);
+out:
         cl_io_fini(env, io);
-        RETURN(result);
+
+        if (iot == CIT_READ) {
+                if (result >= 0)
+                        ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
+                                           LPROC_LL_READ_BYTES, result);
+        } else if (iot == CIT_WRITE) {
+                if (result >= 0) {
+                        ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
+                                           LPROC_LL_WRITE_BYTES, result);
+                        lli->lli_write_rc = 0;
+                } else {
+                        lli->lli_write_rc = result;
+                }
+        }
+
+        return result;
 }
 
 
@@ -886,7 +946,7 @@ static ssize_t ll_file_readv(struct file *file, const struct iovec *iov,
                               unsigned long nr_segs, loff_t *ppos)
 {
         struct lu_env      *env;
-        struct ccc_io_args *args;
+        struct vvp_io_args *args;
         size_t              count;
         ssize_t             result;
         int                 refcheck;
@@ -900,10 +960,10 @@ static ssize_t ll_file_readv(struct file *file, const struct iovec *iov,
         if (IS_ERR(env))
                 RETURN(PTR_ERR(env));
 
-        args = &vvp_env_info(env)->vti_args;
-        args->cia_is_sendfile = 0;
-        args->cia_iov = (struct iovec *)iov;
-        args->cia_nrsegs = nr_segs;
+        args = vvp_env_args(env, IO_NORMAL);
+        args->u.normal.via_iov = (struct iovec *)iov;
+        args->u.normal.via_nrsegs = nr_segs;
+
         result = ll_file_io_generic(env, args, file, CIT_READ, ppos, count);
         cl_env_put(env, &refcheck);
         RETURN(result);
@@ -935,7 +995,7 @@ static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
                                 unsigned long nr_segs, loff_t pos)
 {
         struct lu_env      *env;
-        struct ccc_io_args *args;
+        struct vvp_io_args *args;
         size_t              count;
         ssize_t             result;
         int                 refcheck;
@@ -949,11 +1009,11 @@ static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
         if (IS_ERR(env))
                 RETURN(PTR_ERR(env));
 
-        args = &vvp_env_info(env)->vti_args;
-        args->cia_is_sendfile = 0;
-        args->cia_iov = (struct iovec *)iov;
-        args->cia_nrsegs = nr_segs;
-        args->cia_iocb = iocb;
+        args = vvp_env_args(env, IO_NORMAL);
+        args->u.normal.via_iov = (struct iovec *)iov;
+        args->u.normal.via_nrsegs = nr_segs;
+        args->u.normal.via_iocb = iocb;
+
         result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
                                     &iocb->ki_pos, count);
         cl_env_put(env, &refcheck);
@@ -998,7 +1058,7 @@ static ssize_t ll_file_writev(struct file *file, const struct iovec *iov,
                               unsigned long nr_segs, loff_t *ppos)
 {
         struct lu_env      *env;
-        struct ccc_io_args *args;
+        struct vvp_io_args *args;
         size_t              count;
         ssize_t             result;
         int                 refcheck;
@@ -1012,9 +1072,10 @@ static ssize_t ll_file_writev(struct file *file, const struct iovec *iov,
         if (IS_ERR(env))
                 RETURN(PTR_ERR(env));
 
-        args = &vvp_env_info(env)->vti_args;
-        args->cia_iov = (struct iovec *)iov;
-        args->cia_nrsegs = nr_segs;
+        args = vvp_env_args(env, IO_NORMAL);
+        args->u.normal.via_iov = (struct iovec *)iov;
+        args->u.normal.via_nrsegs = nr_segs;
+
         result = ll_file_io_generic(env, args, file, CIT_WRITE, ppos, count);
         cl_env_put(env, &refcheck);
         RETURN(result);
@@ -1047,7 +1108,7 @@ static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
                                  unsigned long nr_segs, loff_t pos)
 {
         struct lu_env      *env;
-        struct ccc_io_args *args;
+        struct vvp_io_args *args;
         size_t              count;
         ssize_t             result;
         int                 refcheck;
@@ -1061,10 +1122,11 @@ static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
         if (IS_ERR(env))
                 RETURN(PTR_ERR(env));
 
-        args = &vvp_env_info(env)->vti_args;
-        args->cia_iov = (struct iovec *)iov;
-        args->cia_nrsegs = nr_segs;
-        args->cia_iocb = iocb;
+        args = vvp_env_args(env, IO_NORMAL);
+        args->u.normal.via_iov = (struct iovec *)iov;
+        args->u.normal.via_nrsegs = nr_segs;
+        args->u.normal.via_iocb = iocb;
+
         result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
                                   &iocb->ki_pos, count);
         cl_env_put(env, &refcheck);
@@ -1102,6 +1164,7 @@ static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
 #endif
 
 
+#ifdef HAVE_KERNEL_SENDFILE
 /*
  * Send file content (through pagecache) somewhere with helper
  */
@@ -1109,7 +1172,7 @@ static ssize_t ll_file_sendfile(struct file *in_file, loff_t *ppos,size_t count,
                                 read_actor_t actor, void *target)
 {
         struct lu_env      *env;
-        struct ccc_io_args *args;
+        struct vvp_io_args *args;
         ssize_t             result;
         int                 refcheck;
         ENTRY;
@@ -1118,20 +1181,48 @@ static ssize_t ll_file_sendfile(struct file *in_file, loff_t *ppos,size_t count,
         if (IS_ERR(env))
                 RETURN(PTR_ERR(env));
 
-        args = &vvp_env_info(env)->vti_args;
-        args->cia_is_sendfile = 1;
-        args->cia_target = target;
-        args->cia_actor = actor;
+        args = vvp_env_args(env, IO_SENDFILE);
+        args->u.sendfile.via_target = target;
+        args->u.sendfile.via_actor = actor;
+
         result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
         cl_env_put(env, &refcheck);
         RETURN(result);
 }
+#endif
 
-static int ll_lov_recreate_obj(struct inode *inode, struct file *file,
-                               unsigned long arg)
+#ifdef HAVE_KERNEL_SPLICE_READ
+/*
+ * Send file content (through pagecache) somewhere with helper
+ */
+static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
+                                   struct pipe_inode_info *pipe, size_t count,
+                                   unsigned int flags)
+{
+        struct lu_env      *env;
+        struct vvp_io_args *args;
+        ssize_t             result;
+        int                 refcheck;
+        ENTRY;
+
+        env = cl_env_get(&refcheck);
+        if (IS_ERR(env))
+                RETURN(PTR_ERR(env));
+
+        args = vvp_env_args(env, IO_SPLICE);
+        args->u.splice.via_pipe = pipe;
+        args->u.splice.via_flags = flags;
+
+        result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
+        cl_env_put(env, &refcheck);
+        RETURN(result);
+}
+#endif
+
+static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq,
+                           obd_count ost_idx)
 {
         struct obd_export *exp = ll_i2dtexp(inode);
-        struct ll_recreate_obj ucreatp;
         struct obd_trans_info oti = { 0 };
         struct obdo *oa = NULL;
         int lsm_size;
@@ -1139,13 +1230,6 @@ static int ll_lov_recreate_obj(struct inode *inode, struct file *file,
         struct lov_stripe_md *lsm, *lsm2;
         ENTRY;
 
-        if (!cfs_capable(CFS_CAP_SYS_ADMIN))
-                RETURN(-EPERM);
-
-        if (copy_from_user(&ucreatp, (struct ll_recreate_obj *)arg,
-                           sizeof(struct ll_recreate_obj)))
-                RETURN(-EFAULT);
-
         OBDO_ALLOC(oa);
         if (oa == NULL)
                 RETURN(-ENOMEM);
@@ -1157,22 +1241,21 @@ static int ll_lov_recreate_obj(struct inode *inode, struct file *file,
         lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
                    (lsm->lsm_stripe_count));
 
-        OBD_ALLOC(lsm2, lsm_size);
+        OBD_ALLOC_LARGE(lsm2, lsm_size);
         if (lsm2 == NULL)
                 GOTO(out, rc = -ENOMEM);
 
-        oa->o_id = ucreatp.lrc_id;
-        oa->o_gr = ucreatp.lrc_group;
-        oa->o_nlink = ucreatp.lrc_ost_idx;
+        oa->o_id = id;
+        oa->o_seq = seq;
+        oa->o_nlink = ost_idx;
         oa->o_flags |= OBD_FL_RECREATE_OBJS;
         oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
-        obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
-                        OBD_MD_FLMTIME | OBD_MD_FLCTIME);
-
+        obdo_from_inode(oa, inode, &ll_i2info(inode)->lli_fid, OBD_MD_FLTYPE |
+                        OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME);
         memcpy(lsm2, lsm, lsm_size);
         rc = obd_create(exp, oa, &lsm2, &oti);
 
-        OBD_FREE(lsm2, lsm_size);
+        OBD_FREE_LARGE(lsm2, lsm_size);
         GOTO(out, rc);
 out:
         ll_inode_size_unlock(inode, 0);
@@ -1180,6 +1263,41 @@ out:
         return rc;
 }
 
+static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
+{
+        struct ll_recreate_obj ucreat;
+        ENTRY;
+
+        if (!cfs_capable(CFS_CAP_SYS_ADMIN))
+                RETURN(-EPERM);
+
+        if (cfs_copy_from_user(&ucreat, (struct ll_recreate_obj *)arg,
+                               sizeof(struct ll_recreate_obj)))
+                RETURN(-EFAULT);
+
+        RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0,
+                               ucreat.lrc_ost_idx));
+}
+
+static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
+{
+        struct lu_fid fid;
+        obd_id id;
+        obd_count ost_idx;
+        ENTRY;
+
+        if (!cfs_capable(CFS_CAP_SYS_ADMIN))
+                RETURN(-EPERM);
+
+        if (cfs_copy_from_user(&fid, (struct lu_fid *)arg,
+                               sizeof(struct lu_fid)))
+                RETURN(-EFAULT);
+
+        id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32);
+        ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
+        RETURN(ll_lov_recreate(inode, id, 0, ost_idx));
+}
+
 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
                              int flags, struct lov_user_md *lum, int lum_size)
 {
@@ -1225,19 +1343,22 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
         struct mdt_body  *body;
         struct lov_mds_md *lmm = NULL;
         struct ptlrpc_request *req = NULL;
-        struct obd_capa *oc;
+        struct md_op_data *op_data;
         int rc, lmmsize;
 
         rc = ll_get_max_mdsize(sbi, &lmmsize);
         if (rc)
                 RETURN(rc);
 
-        oc = ll_mdscapa_get(inode);
-        rc = md_getattr_name(sbi->ll_md_exp, ll_inode2fid(inode),
-                             oc, filename, strlen(filename) + 1,
-                             OBD_MD_FLEASIZE | OBD_MD_FLDIREA, lmmsize,
-                             ll_i2suppgid(inode), &req);
-        capa_put(oc);
+        op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
+                                     strlen(filename), lmmsize,
+                                     LUSTRE_OPC_ANY, NULL);
+        if (op_data == NULL)
+                RETURN(-ENOMEM);
+
+        op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
+        rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
+        ll_finish_md_op_data(op_data);
         if (rc < 0) {
                 CDEBUG(D_INFO, "md_getattr_name failed "
                        "on %s: rc %d\n", filename, rc);
@@ -1258,8 +1379,7 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
         LASSERT(lmm != NULL);
 
         if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
-            (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3)) &&
-            (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_JOIN))) {
+            (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
                 GOTO(out, rc = -EPROTO);
         }
 
@@ -1283,62 +1403,9 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
                                 lustre_swab_lov_user_md_objects(
                                  ((struct lov_user_md_v3 *)lmm)->lmm_objects,
                                  ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
-                } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_JOIN)) {
-                        lustre_swab_lov_user_md_join((struct lov_user_md_join *)lmm);
                 }
         }
 
-        if (lmm->lmm_magic == LOV_MAGIC_JOIN) {
-                struct lov_stripe_md *lsm;
-                struct lov_user_md_join *lmj;
-                int lmj_size, i, aindex = 0;
-
-                rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
-                if (rc < 0)
-                        GOTO(out, rc = -ENOMEM);
-                rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp, lsm);
-                if (rc)
-                        GOTO(out_free_memmd, rc);
-
-                lmj_size = sizeof(struct lov_user_md_join) +
-                           lsm->lsm_stripe_count *
-                           sizeof(struct lov_user_ost_data_join);
-                OBD_ALLOC(lmj, lmj_size);
-                if (!lmj)
-                        GOTO(out_free_memmd, rc = -ENOMEM);
-
-                memcpy(lmj, lmm, sizeof(struct lov_user_md_join));
-                for (i = 0; i < lsm->lsm_stripe_count; i++) {
-                        struct lov_extent *lex =
-                                &lsm->lsm_array->lai_ext_array[aindex];
-
-                        if (lex->le_loi_idx + lex->le_stripe_count <= i)
-                                aindex ++;
-                        CDEBUG(D_INFO, "aindex %d i %d l_extent_start "
-                                        LPU64" len %d\n", aindex, i,
-                                        lex->le_start, (int)lex->le_len);
-                        lmj->lmm_objects[i].l_extent_start =
-                                lex->le_start;
-
-                        if ((int)lex->le_len == -1)
-                                lmj->lmm_objects[i].l_extent_end = -1;
-                        else
-                                lmj->lmm_objects[i].l_extent_end =
-                                        lex->le_start + lex->le_len;
-                        lmj->lmm_objects[i].l_object_id =
-                                lsm->lsm_oinfo[i]->loi_id;
-                        lmj->lmm_objects[i].l_object_gr =
-                                lsm->lsm_oinfo[i]->loi_gr;
-                        lmj->lmm_objects[i].l_ost_gen =
-                                lsm->lsm_oinfo[i]->loi_ost_gen;
-                        lmj->lmm_objects[i].l_ost_idx =
-                                lsm->lsm_oinfo[i]->loi_ost_idx;
-                }
-                lmm = (struct lov_mds_md *)lmj;
-                lmmsize = lmj_size;
-out_free_memmd:
-                obd_free_memmd(sbi->ll_dt_exp, &lsm);
-        }
 out:
         *lmmp = lmm;
         *lmm_size = lmmsize;
@@ -1359,18 +1426,18 @@ static int ll_lov_setea(struct inode *inode, struct file *file,
         if (!cfs_capable(CFS_CAP_SYS_ADMIN))
                 RETURN(-EPERM);
 
-        OBD_ALLOC(lump, lum_size);
+        OBD_ALLOC_LARGE(lump, lum_size);
         if (lump == NULL) {
                 RETURN(-ENOMEM);
         }
-        if (copy_from_user(lump, (struct lov_user_md  *)arg, lum_size)) {
-                OBD_FREE(lump, lum_size);
+        if (cfs_copy_from_user(lump, (struct lov_user_md  *)arg, lum_size)) {
+                OBD_FREE_LARGE(lump, lum_size);
                 RETURN(-EFAULT);
         }
 
         rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
 
-        OBD_FREE(lump, lum_size);
+        OBD_FREE_LARGE(lump, lum_size);
         RETURN(rc);
 }
 
@@ -1388,12 +1455,12 @@ static int ll_lov_setstripe(struct inode *inode, struct file *file,
 
         /* first try with v1 which is smaller than v3 */
         lum_size = sizeof(struct lov_user_md_v1);
-        if (copy_from_user(lumv1, lumv1p, lum_size))
+        if (cfs_copy_from_user(lumv1, lumv1p, lum_size))
                 RETURN(-EFAULT);
 
         if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
                 lum_size = sizeof(struct lov_user_md_v3);
-                if (copy_from_user(&lumv3, lumv3p, lum_size))
+                if (cfs_copy_from_user(&lumv3, lumv3p, lum_size))
                         RETURN(-EFAULT);
         }
 
@@ -1410,12 +1477,13 @@ static int ll_lov_setstripe(struct inode *inode, struct file *file,
 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
 {
         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
+        int rc = -ENODATA;
+        ENTRY;
 
-        if (!lsm)
-                RETURN(-ENODATA);
-
-        return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, lsm,
-                            (void *)arg);
+        if (lsm != NULL)
+                rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
+                                   lsm, (void *)arg);
+        RETURN(rc);
 }
 
 int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
@@ -1426,32 +1494,35 @@ int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
         int                     rc;
         ENTRY;
 
-        spin_lock(&lli->lli_lock);
+        if (ll_file_nolock(file))
+                RETURN(-EOPNOTSUPP);
+
+        cfs_spin_lock(&lli->lli_lock);
         if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
-                CERROR("group lock already existed with gid %lu\n",
+                CWARN("group lock already existed with gid %lu\n",
                        fd->fd_grouplock.cg_gid);
-                spin_unlock(&lli->lli_lock);
+                cfs_spin_unlock(&lli->lli_lock);
                 RETURN(-EINVAL);
         }
         LASSERT(fd->fd_grouplock.cg_lock == NULL);
-        spin_unlock(&lli->lli_lock);
+        cfs_spin_unlock(&lli->lli_lock);
 
         rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
                               arg, (file->f_flags & O_NONBLOCK), &grouplock);
         if (rc)
                 RETURN(rc);
 
-        spin_lock(&lli->lli_lock);
+        cfs_spin_lock(&lli->lli_lock);
         if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
-                spin_unlock(&lli->lli_lock);
+                cfs_spin_unlock(&lli->lli_lock);
                 CERROR("another thread just won the race\n");
                 cl_put_grouplock(&grouplock);
                 RETURN(-EINVAL);
         }
 
-        fd->fd_flags |= (LL_FILE_GROUP_LOCKED | LL_FILE_IGNORE_LOCK);
+        fd->fd_flags |= LL_FILE_GROUP_LOCKED;
         fd->fd_grouplock = grouplock;
-        spin_unlock(&lli->lli_lock);
+        cfs_spin_unlock(&lli->lli_lock);
 
         CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
         RETURN(0);
@@ -1464,206 +1535,31 @@ int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg)
         struct ccc_grouplock    grouplock;
         ENTRY;
 
-        spin_lock(&lli->lli_lock);
+        cfs_spin_lock(&lli->lli_lock);
         if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
-                spin_unlock(&lli->lli_lock);
-                CERROR("no group lock held\n");
+                cfs_spin_unlock(&lli->lli_lock);
+                CWARN("no group lock held\n");
                 RETURN(-EINVAL);
         }
         LASSERT(fd->fd_grouplock.cg_lock != NULL);
 
         if (fd->fd_grouplock.cg_gid != arg) {
-                CERROR("group lock %lu doesn't match current id %lu\n",
+                CWARN("group lock %lu doesn't match current id %lu\n",
                        arg, fd->fd_grouplock.cg_gid);
-                spin_unlock(&lli->lli_lock);
+                cfs_spin_unlock(&lli->lli_lock);
                 RETURN(-EINVAL);
         }
 
         grouplock = fd->fd_grouplock;
-        fd->fd_grouplock.cg_env = NULL;
-        fd->fd_grouplock.cg_lock = NULL;
-        fd->fd_grouplock.cg_gid = 0;
-        fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED | LL_FILE_IGNORE_LOCK);
-        spin_unlock(&lli->lli_lock);
+        memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
+        fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
+        cfs_spin_unlock(&lli->lli_lock);
 
         cl_put_grouplock(&grouplock);
         CDEBUG(D_INFO, "group lock %lu released\n", arg);
         RETURN(0);
 }
 
-#if LUSTRE_FIX >= 50
-static int join_sanity_check(struct inode *head, struct inode *tail)
-{
-        ENTRY;
-        if ((ll_i2sbi(head)->ll_flags & LL_SBI_JOIN) == 0) {
-                CERROR("server do not support join \n");
-                RETURN(-EINVAL);
-        }
-        if (!S_ISREG(tail->i_mode) || !S_ISREG(head->i_mode)) {
-                CERROR("tail ino %lu and ino head %lu must be regular\n",
-                       head->i_ino, tail->i_ino);
-                RETURN(-EINVAL);
-        }
-        if (head->i_ino == tail->i_ino) {
-                CERROR("file %lu can not be joined to itself \n", head->i_ino);
-                RETURN(-EINVAL);
-        }
-        if (i_size_read(head) % JOIN_FILE_ALIGN) {
-                CERROR("hsize %llu must be times of 64K\n", i_size_read(head));
-                RETURN(-EINVAL);
-        }
-        RETURN(0);
-}
-
-static int join_file(struct inode *head_inode, struct file *head_filp,
-                     struct file *tail_filp)
-{
-        struct dentry *tail_dentry = tail_filp->f_dentry;
-        struct lookup_intent oit = {.it_op = IT_OPEN,
-                                    .it_flags = head_filp->f_flags,
-                                    .it_create_mode = M_JOIN_FILE};
-        struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CW,
-                ll_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
-
-        struct lustre_handle lockh;
-        struct md_op_data *op_data;
-        int    rc;
-        loff_t data;
-        ENTRY;
-
-        tail_dentry = tail_filp->f_dentry;
-
-        data = i_size_read(head_inode);
-        op_data = ll_prep_md_op_data(NULL, head_inode,
-                                     tail_dentry->d_parent->d_inode,
-                                     tail_dentry->d_name.name,
-                                     tail_dentry->d_name.len, 0,
-                                     LUSTRE_OPC_ANY, &data);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
-
-        rc = md_enqueue(ll_i2mdexp(head_inode), &einfo, &oit,
-                         op_data, &lockh, NULL, 0, NULL, 0);
-
-        ll_finish_md_op_data(op_data);
-        if (rc < 0)
-                GOTO(out, rc);
-
-        rc = oit.d.lustre.it_status;
-
-        if (rc < 0 || it_open_error(DISP_OPEN_OPEN, &oit)) {
-                rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, &oit);
-                ptlrpc_req_finished((struct ptlrpc_request *)
-                                    oit.d.lustre.it_data);
-                GOTO(out, rc);
-        }
-
-        if (oit.d.lustre.it_lock_mode) { /* If we got lock - release it right
-                                           * away */
-                ldlm_lock_decref(&lockh, oit.d.lustre.it_lock_mode);
-                oit.d.lustre.it_lock_mode = 0;
-        }
-        ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data);
-        it_clear_disposition(&oit, DISP_ENQ_COMPLETE);
-        ll_release_openhandle(head_filp->f_dentry, &oit);
-out:
-        ll_intent_release(&oit);
-        RETURN(rc);
-}
-
-static int ll_file_join(struct inode *head, struct file *filp,
-                        char *filename_tail)
-{
-        struct inode *tail = NULL, *first = NULL, *second = NULL;
-        struct dentry *tail_dentry;
-        struct file *tail_filp, *first_filp, *second_filp;
-        struct ll_lock_tree first_tree, second_tree;
-        struct ll_lock_tree_node *first_node, *second_node;
-        struct ll_inode_info *hlli = ll_i2info(head);
-        int rc = 0, cleanup_phase = 0;
-        ENTRY;
-
-        CDEBUG(D_VFSTRACE, "VFS Op:head=%lu/%u(%p) tail %s\n",
-               head->i_ino, head->i_generation, head, filename_tail);
-
-        tail_filp = filp_open(filename_tail, O_WRONLY, 0644);
-        if (IS_ERR(tail_filp)) {
-                CERROR("Can not open tail file %s", filename_tail);
-                rc = PTR_ERR(tail_filp);
-                GOTO(cleanup, rc);
-        }
-        tail = igrab(tail_filp->f_dentry->d_inode);
-
-        tail_dentry = tail_filp->f_dentry;
-        LASSERT(tail_dentry);
-        cleanup_phase = 1;
-
-        /*reorder the inode for lock sequence*/
-        first = head->i_ino > tail->i_ino ? head : tail;
-        second = head->i_ino > tail->i_ino ? tail : head;
-        first_filp = head->i_ino > tail->i_ino ? filp : tail_filp;
-        second_filp = head->i_ino > tail->i_ino ? tail_filp : filp;
-
-        CDEBUG(D_INFO, "reorder object from %lu:%lu to %lu:%lu \n",
-               head->i_ino, tail->i_ino, first->i_ino, second->i_ino);
-        first_node = ll_node_from_inode(first, 0, OBD_OBJECT_EOF, LCK_EX);
-        if (IS_ERR(first_node)){
-                rc = PTR_ERR(first_node);
-                GOTO(cleanup, rc);
-        }
-        first_tree.lt_fd = first_filp->private_data;
-        rc = ll_tree_lock(&first_tree, first_node, NULL, 0, 0);
-        if (rc != 0)
-                GOTO(cleanup, rc);
-        cleanup_phase = 2;
-
-        second_node = ll_node_from_inode(second, 0, OBD_OBJECT_EOF, LCK_EX);
-        if (IS_ERR(second_node)){
-                rc = PTR_ERR(second_node);
-                GOTO(cleanup, rc);
-        }
-        second_tree.lt_fd = second_filp->private_data;
-        rc = ll_tree_lock(&second_tree, second_node, NULL, 0, 0);
-        if (rc != 0)
-                GOTO(cleanup, rc);
-        cleanup_phase = 3;
-
-        rc = join_sanity_check(head, tail);
-        if (rc)
-                GOTO(cleanup, rc);
-
-        rc = join_file(head, filp, tail_filp);
-        if (rc)
-                GOTO(cleanup, rc);
-cleanup:
-        switch (cleanup_phase) {
-        case 3:
-                ll_tree_unlock(&second_tree);
-                obd_cancel_unused(ll_i2dtexp(second),
-                                  ll_i2info(second)->lli_smd, 0, NULL);
-        case 2:
-                ll_tree_unlock(&first_tree);
-                obd_cancel_unused(ll_i2dtexp(first),
-                                  ll_i2info(first)->lli_smd, 0, NULL);
-        case 1:
-                filp_close(tail_filp, 0);
-                if (tail)
-                        iput(tail);
-                if (head && rc == 0) {
-                        obd_free_memmd(ll_i2sbi(head)->ll_dt_exp,
-                                       &hlli->lli_smd);
-                        hlli->lli_smd = NULL;
-                }
-        case 0:
-                break;
-        default:
-                CERROR("invalid cleanup_phase %d\n", cleanup_phase);
-                LBUG();
-        }
-        RETURN(rc);
-}
-#endif /* LUSTRE_FIX >= 50 */
-
 /**
  * Close inode open handle
  *
@@ -1713,7 +1609,7 @@ int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
  * Get size for inode for which FIEMAP mapping is requested.
  * Make the FIEMAP get_info call and returns the result.
  */
-int ll_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
+int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
               int num_bytes)
 {
         struct obd_export *exp = ll_i2dtexp(inode);
@@ -1723,6 +1619,19 @@ int ll_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
         int rc;
         ENTRY;
 
+        /* Checks for fiemap flags */
+        if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
+                fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
+                return -EBADR;
+        }
+
+        /* Check for FIEMAP_FLAG_SYNC */
+        if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
+                rc = filemap_fdatawrite(inode->i_mapping);
+                if (rc)
+                        return rc;
+        }
+
         /* If the stripe_count > 1 and the application does not understand
          * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
          */
@@ -1731,12 +1640,11 @@ int ll_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
                 return -EOPNOTSUPP;
 
         fm_key.oa.o_id = lsm->lsm_object_id;
-        fm_key.oa.o_gr = lsm->lsm_object_gr;
+        fm_key.oa.o_seq = lsm->lsm_object_seq;
         fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
 
-        obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLFID | OBD_MD_FLGROUP |
+        obdo_from_inode(&fm_key.oa, inode, &ll_i2info(inode)->lli_fid,
                         OBD_MD_FLSIZE);
-
         /* If filesize is 0, then there would be no objects for mapping */
         if (fm_key.oa.o_size == 0) {
                 fiemap->fm_mapped_extents = 0;
@@ -1752,9 +1660,103 @@ int ll_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
         RETURN(rc);
 }
 
+int ll_fid2path(struct obd_export *exp, void *arg)
+{
+        struct getinfo_fid2path *gfout, *gfin;
+        int outsize, rc;
+        ENTRY;
+
+        /* Need to get the buflen */
+        OBD_ALLOC_PTR(gfin);
+        if (gfin == NULL)
+                RETURN(-ENOMEM);
+        if (cfs_copy_from_user(gfin, arg, sizeof(*gfin))) {
+                OBD_FREE_PTR(gfin);
+                RETURN(-EFAULT);
+        }
+
+        outsize = sizeof(*gfout) + gfin->gf_pathlen;
+        OBD_ALLOC(gfout, outsize);
+        if (gfout == NULL) {
+                OBD_FREE_PTR(gfin);
+                RETURN(-ENOMEM);
+        }
+        memcpy(gfout, gfin, sizeof(*gfout));
+        OBD_FREE_PTR(gfin);
+
+        /* Call mdc_iocontrol */
+        rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
+        if (rc)
+                GOTO(gf_free, rc);
+        if (cfs_copy_to_user(arg, gfout, outsize))
+                rc = -EFAULT;
+
+gf_free:
+        OBD_FREE(gfout, outsize);
+        RETURN(rc);
+}
+
+static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
+{
+        struct ll_user_fiemap *fiemap_s;
+        size_t num_bytes, ret_bytes;
+        unsigned int extent_count;
+        int rc = 0;
+
+        /* Get the extent count so we can calculate the size of
+         * required fiemap buffer */
+        if (get_user(extent_count,
+            &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
+                RETURN(-EFAULT);
+        num_bytes = sizeof(*fiemap_s) + (extent_count *
+                                         sizeof(struct ll_fiemap_extent));
+
+        OBD_ALLOC_LARGE(fiemap_s, num_bytes);
+        if (fiemap_s == NULL)
+                RETURN(-ENOMEM);
+
+        /* get the fiemap value */
+        if (copy_from_user(fiemap_s,(struct ll_user_fiemap __user *)arg,
+                           sizeof(*fiemap_s)))
+                GOTO(error, rc = -EFAULT);
+
+        /* If fm_extent_count is non-zero, read the first extent since
+         * it is used to calculate end_offset and device from previous
+         * fiemap call. */
+        if (extent_count) {
+                if (copy_from_user(&fiemap_s->fm_extents[0],
+                    (char __user *)arg + sizeof(*fiemap_s),
+                    sizeof(struct ll_fiemap_extent)))
+                        GOTO(error, rc = -EFAULT);
+        }
+
+        rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
+        if (rc)
+                GOTO(error, rc);
+
+        ret_bytes = sizeof(struct ll_user_fiemap);
+
+        if (extent_count != 0)
+                ret_bytes += (fiemap_s->fm_mapped_extents *
+                                 sizeof(struct ll_fiemap_extent));
+
+        if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
+                rc = -EFAULT;
+
+error:
+        OBD_FREE_LARGE(fiemap_s, num_bytes);
+        RETURN(rc);
+}
+
+#ifdef HAVE_UNLOCKED_IOCTL
+long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+        struct inode *inode = file->f_dentry->d_inode;
+#else
 int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
                   unsigned long arg)
 {
+#endif
         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
         int flags;
         ENTRY;
@@ -1800,96 +1802,17 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
         case LL_IOC_LOV_GETSTRIPE:
                 RETURN(ll_lov_getstripe(inode, arg));
         case LL_IOC_RECREATE_OBJ:
-                RETURN(ll_lov_recreate_obj(inode, file, arg));
-        case EXT3_IOC_FIEMAP: {
-                struct ll_user_fiemap *fiemap_s;
-                size_t num_bytes, ret_bytes;
-                unsigned int extent_count;
-                int rc = 0;
-
-                /* Get the extent count so we can calculate the size of
-                 * required fiemap buffer */
-                if (get_user(extent_count,
-                    &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
-                        RETURN(-EFAULT);
-                num_bytes = sizeof(*fiemap_s) + (extent_count *
-                                                 sizeof(struct ll_fiemap_extent));
-                OBD_VMALLOC(fiemap_s, num_bytes);
-                if (fiemap_s == NULL)
-                        RETURN(-ENOMEM);
-
-                if (copy_from_user(fiemap_s,(struct ll_user_fiemap __user *)arg,
-                                   sizeof(*fiemap_s)))
-                        GOTO(error, rc = -EFAULT);
-
-                if (fiemap_s->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
-                        fiemap_s->fm_flags = fiemap_s->fm_flags &
-                                                    ~LUSTRE_FIEMAP_FLAGS_COMPAT;
-                        if (copy_to_user((char *)arg, fiemap_s,
-                                         sizeof(*fiemap_s)))
-                                GOTO(error, rc = -EFAULT);
-
-                        GOTO(error, rc = -EBADR);
-                }
-
-                /* If fm_extent_count is non-zero, read the first extent since
-                 * it is used to calculate end_offset and device from previous
-                 * fiemap call. */
-                if (extent_count) {
-                        if (copy_from_user(&fiemap_s->fm_extents[0],
-                            (char __user *)arg + sizeof(*fiemap_s),
-                            sizeof(struct ll_fiemap_extent)))
-                                GOTO(error, rc = -EFAULT);
-                }
-
-                if (fiemap_s->fm_flags & FIEMAP_FLAG_SYNC) {
-                        int rc;
-
-                        rc = filemap_fdatawrite(inode->i_mapping);
-                        if (rc)
-                                GOTO(error, rc);
-                }
-
-                rc = ll_fiemap(inode, fiemap_s, num_bytes);
-                if (rc)
-                        GOTO(error, rc);
-
-                ret_bytes = sizeof(struct ll_user_fiemap);
-
-                if (extent_count != 0)
-                        ret_bytes += (fiemap_s->fm_mapped_extents *
-                                         sizeof(struct ll_fiemap_extent));
-
-                if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
-                        rc = -EFAULT;
-
-error:
-                OBD_VFREE(fiemap_s, num_bytes);
-                RETURN(rc);
-        }
-        case EXT3_IOC_GETFLAGS:
-        case EXT3_IOC_SETFLAGS:
+                RETURN(ll_lov_recreate_obj(inode, arg));
+        case LL_IOC_RECREATE_FID:
+                RETURN(ll_lov_recreate_fid(inode, arg));
+        case FSFILT_IOC_FIEMAP:
+                RETURN(ll_ioctl_fiemap(inode, arg));
+        case FSFILT_IOC_GETFLAGS:
+        case FSFILT_IOC_SETFLAGS:
                 RETURN(ll_iocontrol(inode, file, cmd, arg));
-        case EXT3_IOC_GETVERSION_OLD:
-        case EXT3_IOC_GETVERSION:
+        case FSFILT_IOC_GETVERSION_OLD:
+        case FSFILT_IOC_GETVERSION:
                 RETURN(put_user(inode->i_generation, (int *)arg));
-        case LL_IOC_JOIN: {
-#if LUSTRE_FIX >= 50
-                /* Allow file join in beta builds to allow debuggging */
-                char *ftail;
-                int rc;
-
-                ftail = getname((const char *)arg);
-                if (IS_ERR(ftail))
-                        RETURN(PTR_ERR(ftail));
-                rc = ll_file_join(inode, file, ftail);
-                putname(ftail);
-                RETURN(rc);
-#else
-                CWARN("file join is not supported in this version of Lustre\n");
-                RETURN(-ENOTTY);
-#endif
-        }
         case LL_IOC_GROUP_LOCK:
                 RETURN(ll_get_grouplock(inode, file, arg));
         case LL_IOC_GROUP_UNLOCK:
@@ -1900,18 +1823,34 @@ error:
         /* We need to special case any other ioctls we want to handle,
          * to send them to the MDS/OST as appropriate and to properly
          * network encode the arg field.
-        case EXT3_IOC_SETVERSION_OLD:
-        case EXT3_IOC_SETVERSION:
+        case FSFILT_IOC_SETVERSION_OLD:
+        case FSFILT_IOC_SETVERSION:
         */
         case LL_IOC_FLUSHCTX:
                 RETURN(ll_flush_ctx(inode));
         case LL_IOC_PATH2FID: {
-                if (copy_to_user((void *)arg, &ll_i2info(inode)->lli_fid,
-                                 sizeof(struct lu_fid)))
+                if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
+                                     sizeof(struct lu_fid)))
                         RETURN(-EFAULT);
 
                 RETURN(0);
         }
+        case OBD_IOC_FID2PATH:
+                RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
+
+        case LL_IOC_GET_MDTIDX: {
+                int mdtidx;
+
+                mdtidx = ll_get_mdt_idx(inode);
+                if (mdtidx < 0)
+                        RETURN(mdtidx);
+
+                if (put_user((int)mdtidx, (int*)arg))
+                        RETURN(-EFAULT);
+
+                RETURN(0);
+        }
+
         default: {
                 int err;
 
@@ -1947,9 +1886,7 @@ loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
                 if (rc != 0)
                         RETURN(rc);
 
-                ll_inode_size_lock(inode, 0);
                 offset += i_size_read(inode);
-                ll_inode_size_unlock(inode, 0);
         } else if (origin == 1) { /* SEEK_CUR */
                 offset += file->f_pos;
         }
@@ -1965,6 +1902,34 @@ loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
         RETURN(retval);
 }
 
+#ifdef HAVE_FLUSH_OWNER_ID
+int ll_flush(struct file *file, fl_owner_t id)
+#else
+int ll_flush(struct file *file)
+#endif
+{
+        struct inode *inode = file->f_dentry->d_inode;
+        struct ll_inode_info *lli = ll_i2info(inode);
+        struct lov_stripe_md *lsm = lli->lli_smd;
+        int rc, err;
+
+        /* the application should know write failure already. */
+        if (lli->lli_write_rc)
+                return 0;
+
+        /* catch async errors that were recorded back when async writeback
+         * failed for pages in this mapping. */
+        rc = lli->lli_async_rc;
+        lli->lli_async_rc = 0;
+        if (lsm) {
+                err = lov_test_and_clear_async_rc(lsm);
+                if (rc == 0)
+                        rc = err;
+        }
+
+        return rc ? -EIO : 0;
+}
+
 int ll_fsync(struct file *file, struct dentry *dentry, int data)
 {
         struct inode *inode = dentry->d_inode;
@@ -2004,26 +1969,33 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data)
                 ptlrpc_req_finished(req);
 
         if (data && lsm) {
-                struct obdo *oa;
+                struct obd_info *oinfo;
 
-                OBDO_ALLOC(oa);
-                if (!oa)
+                OBD_ALLOC_PTR(oinfo);
+                if (!oinfo)
                         RETURN(rc ? rc : -ENOMEM);
-
-                oa->o_id = lsm->lsm_object_id;
-                oa->o_gr = lsm->lsm_object_gr;
-                oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
-                obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
-                                           OBD_MD_FLMTIME | OBD_MD_FLCTIME |
-                                           OBD_MD_FLGROUP);
-
-                oc = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
-                err = obd_sync(ll_i2sbi(inode)->ll_dt_exp, oa, lsm,
-                               0, OBD_OBJECT_EOF, oc);
-                capa_put(oc);
+                OBDO_ALLOC(oinfo->oi_oa);
+                if (!oinfo->oi_oa) {
+                        OBD_FREE_PTR(oinfo);
+                        RETURN(rc ? rc : -ENOMEM);
+                }
+                oinfo->oi_oa->o_id = lsm->lsm_object_id;
+                oinfo->oi_oa->o_seq = lsm->lsm_object_seq;
+                oinfo->oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
+                obdo_from_inode(oinfo->oi_oa, inode, &ll_i2info(inode)->lli_fid,
+                                OBD_MD_FLTYPE | OBD_MD_FLATIME |
+                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
+                                OBD_MD_FLGROUP);
+                oinfo->oi_md = lsm;
+                oinfo->oi_capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
+                err = obd_sync_rqset(ll_i2sbi(inode)->ll_dt_exp, oinfo, 0,
+                                     OBD_OBJECT_EOF);
+                capa_put(oinfo->oi_capa);
                 if (!rc)
                         rc = err;
-                OBDO_FREE(oa);
+                OBDO_FREE(oinfo->oi_oa);
+                OBD_FREE_PTR(oinfo);
+                lli->lli_write_rc = err < 0 ? : 0;
         }
 
         RETURN(rc);
@@ -2038,7 +2010,7 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
                                            .ei_cbdata = file_lock };
         struct md_op_data *op_data;
         struct lustre_handle lockh = {0};
-        ldlm_policy_data_t flock;
+        ldlm_policy_data_t flock = {{0}};
         int flags = 0;
         int rc;
         ENTRY;
@@ -2050,13 +2022,28 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 
         if (file_lock->fl_flags & FL_FLOCK) {
                 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
-                /* set missing params for flock() calls */
-                file_lock->fl_end = OFFSET_MAX;
-                file_lock->fl_pid = current->tgid;
+                /* flocks are whole-file locks */
+                flock.l_flock.end = OFFSET_MAX;
+                /* For flocks owner is determined by the local file desctiptor*/
+                flock.l_flock.owner = (unsigned long)file_lock->fl_file;
+        } else if (file_lock->fl_flags & FL_POSIX) {
+                flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
+                flock.l_flock.start = file_lock->fl_start;
+                flock.l_flock.end = file_lock->fl_end;
+        } else {
+                RETURN(-EINVAL);
         }
         flock.l_flock.pid = file_lock->fl_pid;
-        flock.l_flock.start = file_lock->fl_start;
-        flock.l_flock.end = file_lock->fl_end;
+
+        /* Somewhat ugly workaround for svc lockd.
+         * lockd installs custom fl_lmops->fl_compare_owner that checks
+         * for the fl_owner to be the same (which it always is on local node
+         * I guess between lockd processes) and then compares pid.
+         * As such we assign pid to the owner field to make it all work,
+         * conflict with normal locks is unlikely since pid space and
+         * pointer space for current->files are not intersecting */
+        if (file_lock->fl_lmops && file_lock->fl_lmops->fl_compare_owner)
+                flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
 
         switch (file_lock->fl_type) {
         case F_RDLCK:
@@ -2142,10 +2129,12 @@ int ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
         RETURN(-ENOSYS);
 }
 
-int ll_have_md_lock(struct inode *inode, __u64 bits)
+int ll_have_md_lock(struct inode *inode, __u64 bits,  ldlm_mode_t l_req_mode)
 {
         struct lustre_handle lockh;
         ldlm_policy_data_t policy = { .l_inodebits = {bits}};
+        ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ?
+                                (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
         struct lu_fid *fid;
         int flags;
         ENTRY;
@@ -2154,11 +2143,12 @@ int ll_have_md_lock(struct inode *inode, __u64 bits)
                RETURN(0);
 
         fid = &ll_i2info(inode)->lli_fid;
-        CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
+        CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
+               ldlm_lockname[mode]);
 
         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
         if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
-                          LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
+                          mode, &lockh)) {
                 RETURN(1);
         }
         RETURN(0);
@@ -2224,10 +2214,16 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
 
         exp = ll_i2mdexp(inode);
 
+        /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
+         *      But under CMD case, it caused some lock issues, should be fixed
+         *      with new CMD ibits lock. See bug 12718 */
         if (exp->exp_connect_flags & OBD_CONNECT_ATTRFID) {
                 struct lookup_intent oit = { .it_op = IT_GETATTR };
                 struct md_op_data *op_data;
 
+                if (ibits == MDS_INODELOCK_LOOKUP)
+                        oit.it_op = IT_LOOKUP;
+
                 /* Call getattr by fid, so do not provide name at all. */
                 op_data = ll_prep_md_op_data(NULL, dentry->d_parent->d_inode,
                                              dentry->d_inode, NULL, 0, 0,
@@ -2259,19 +2255,18 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
                    here to preserve get_cwd functionality on 2.6.
                    Bug 10503 */
                 if (!dentry->d_inode->i_nlink) {
-                        spin_lock(&ll_lookup_lock);
+                        cfs_spin_lock(&ll_lookup_lock);
                         spin_lock(&dcache_lock);
                         ll_drop_dentry(dentry);
                         spin_unlock(&dcache_lock);
-                        spin_unlock(&ll_lookup_lock);
+                        cfs_spin_unlock(&ll_lookup_lock);
                 }
 
                 ll_lookup_finish_locks(&oit, dentry);
-        } else if (!ll_have_md_lock(dentry->d_inode, ibits)) {
-
+        } else if (!ll_have_md_lock(dentry->d_inode, ibits, LCK_MINMODE)) {
                 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
                 obd_valid valid = OBD_MD_FLGETATTR;
-                struct obd_capa *oc;
+                struct md_op_data *op_data;
                 int ealen = 0;
 
                 if (S_ISREG(inode->i_mode)) {
@@ -2280,13 +2275,19 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
                                 RETURN(rc);
                         valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
                 }
+
+                op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
+                                             0, ealen, LUSTRE_OPC_ANY,
+                                             NULL);
+                if (op_data == NULL)
+                        RETURN(-ENOMEM);
+
+                op_data->op_valid = valid;
                 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
                  * capa for this inode. Because we only keep capas of dirs
                  * fresh. */
-                oc = ll_mdscapa_get(inode);
-                rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc, valid,
-                                ealen, &req);
-                capa_put(oc);
+                rc = md_getattr(sbi->ll_md_exp, op_data, &req);
+                ll_finish_md_op_data(op_data);
                 if (rc) {
                         rc = ll_inode_revalidate_fini(inode, rc);
                         RETURN(rc);
@@ -2299,23 +2300,28 @@ out:
         return rc;
 }
 
-int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it)
+int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
+                           __u64 ibits)
 {
+        struct inode *inode = dentry->d_inode;
         int rc;
         ENTRY;
 
-        rc = __ll_inode_revalidate_it(dentry, it, MDS_INODELOCK_UPDATE |
-                                                  MDS_INODELOCK_LOOKUP);
+        rc = __ll_inode_revalidate_it(dentry, it, ibits);
 
         /* if object not yet allocated, don't validate size */
-        if (rc == 0 && ll_i2info(dentry->d_inode)->lli_smd == NULL)
+        if (rc == 0 && ll_i2info(dentry->d_inode)->lli_smd == NULL) {
+                LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime;
+                LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime;
+                LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime;
                 RETURN(0);
+        }
 
         /* cl_glimpse_size will prefer locally cached writes if they extend
          * the file */
 
         if (rc == 0)
-                rc = cl_glimpse_size(dentry->d_inode);
+                rc = cl_glimpse_size(inode);
 
         RETURN(rc);
 }
@@ -2324,16 +2330,22 @@ int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
                   struct lookup_intent *it, struct kstat *stat)
 {
         struct inode *inode = de->d_inode;
+        struct ll_sb_info *sbi = ll_i2sbi(inode);
+        struct ll_inode_info *lli = ll_i2info(inode);
         int res = 0;
 
-        res = ll_inode_revalidate_it(de, it);
-        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETATTR, 1);
+        res = ll_inode_revalidate_it(de, it, MDS_INODELOCK_UPDATE |
+                                             MDS_INODELOCK_LOOKUP);
+        ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
 
         if (res)
                 return res;
 
         stat->dev = inode->i_sb->s_dev;
-        stat->ino = inode->i_ino;
+        if (ll_need_32bit_api(sbi))
+                stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
+        else
+                stat->ino = inode->i_ino;
         stat->mode = inode->i_mode;
         stat->nlink = inode->i_nlink;
         stat->uid = inode->i_uid;
@@ -2348,10 +2360,8 @@ int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
         stat->blksize = 1 << inode->i_blkbits;
 #endif
 
-        ll_inode_size_lock(inode, 0);
         stat->size = i_size_read(inode);
         stat->blocks = inode->i_blocks;
-        ll_inode_size_unlock(inode, 0);
 
         return 0;
 }
@@ -2362,6 +2372,42 @@ int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
         return ll_getattr_it(mnt, de, &it, stat);
 }
 
+#ifdef HAVE_LINUX_FIEMAP_H
+int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
+                __u64 start, __u64 len)
+{
+        int rc;
+        size_t num_bytes;
+        struct ll_user_fiemap *fiemap;
+        unsigned int extent_count = fieinfo->fi_extents_max;
+
+        num_bytes = sizeof(*fiemap) + (extent_count *
+                                       sizeof(struct ll_fiemap_extent));
+        OBD_ALLOC_LARGE(fiemap, num_bytes);
+
+        if (fiemap == NULL)
+                RETURN(-ENOMEM);
+
+        fiemap->fm_flags = fieinfo->fi_flags;
+        fiemap->fm_extent_count = fieinfo->fi_extents_max;
+        fiemap->fm_start = start;
+        fiemap->fm_length = len;
+        memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
+               sizeof(struct ll_fiemap_extent));
+
+        rc = ll_do_fiemap(inode, fiemap, num_bytes);
+
+        fieinfo->fi_flags = fiemap->fm_flags;
+        fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
+        memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
+               fiemap->fm_mapped_extents * sizeof(struct ll_fiemap_extent));
+
+        OBD_FREE_LARGE(fiemap, num_bytes);
+        return rc;
+}
+#endif
+
+
 static
 int lustre_check_acl(struct inode *inode, int mask)
 {
@@ -2371,9 +2417,9 @@ int lustre_check_acl(struct inode *inode, int mask)
         int rc;
         ENTRY;
 
-        spin_lock(&lli->lli_lock);
+        cfs_spin_lock(&lli->lli_lock);
         acl = posix_acl_dup(lli->lli_posix_acl);
-        spin_unlock(&lli->lli_lock);
+        cfs_spin_unlock(&lli->lli_lock);
 
         if (!acl)
                 RETURN(-EAGAIN);
@@ -2388,7 +2434,11 @@ int lustre_check_acl(struct inode *inode, int mask)
 }
 
 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,10))
+#ifndef HAVE_INODE_PERMISION_2ARGS
 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
+#else
+int ll_inode_permission(struct inode *inode, int mask)
+#endif
 {
         int rc = 0;
         ENTRY;
@@ -2397,7 +2447,7 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
         * need to do it before permission check. */
 
         if (inode == inode->i_sb->s_root->d_inode) {
-                struct lookup_intent it = { .it_op = IT_GETATTR };
+                struct lookup_intent it = { .it_op = IT_LOOKUP };
 
                 rc = __ll_inode_revalidate_it(inode->i_sb->s_root, &it,
                                               MDS_INODELOCK_LOOKUP);
@@ -2435,7 +2485,7 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
                 return -EROFS;
         if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
                 return -EACCES;
-        if (current->fsuid == inode->i_uid) {
+        if (cfs_curproc_fsuid() == inode->i_uid) {
                 mode >>= 6;
         } else if (1) {
                 if (((mode >> 3) & mask & S_IRWXO) != mask)
@@ -2448,7 +2498,7 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
                 return rc;
         } else {
 check_groups:
-                if (in_group_p(inode->i_gid))
+                if (cfs_curproc_is_in_groups(inode->i_gid))
                         mode >>= 3;
         }
         if ((mode & mask & S_IRWXO) == mask)
@@ -2486,13 +2536,23 @@ struct file_operations ll_file_operations = {
         .READ_METHOD    = READ_FUNCTION,
         .write          = ll_file_write,
         .WRITE_METHOD   = WRITE_FUNCTION,
+#ifdef HAVE_UNLOCKED_IOCTL
+        .unlocked_ioctl = ll_file_ioctl,
+#else
         .ioctl          = ll_file_ioctl,
+#endif
         .open           = ll_file_open,
         .release        = ll_file_release,
         .mmap           = ll_file_mmap,
         .llseek         = ll_file_seek,
+#ifdef HAVE_KERNEL_SENDFILE
         .sendfile       = ll_file_sendfile,
+#endif
+#ifdef HAVE_KERNEL_SPLICE_READ
+        .splice_read    = ll_file_splice_read,
+#endif
         .fsync          = ll_fsync,
+        .flush          = ll_flush
 };
 
 struct file_operations ll_file_operations_flock = {
@@ -2500,13 +2560,23 @@ struct file_operations ll_file_operations_flock = {
         .READ_METHOD    = READ_FUNCTION,
         .write          = ll_file_write,
         .WRITE_METHOD   = WRITE_FUNCTION,
+#ifdef HAVE_UNLOCKED_IOCTL
+        .unlocked_ioctl = ll_file_ioctl,
+#else
         .ioctl          = ll_file_ioctl,
+#endif
         .open           = ll_file_open,
         .release        = ll_file_release,
         .mmap           = ll_file_mmap,
         .llseek         = ll_file_seek,
+#ifdef HAVE_KERNEL_SENDFILE
         .sendfile       = ll_file_sendfile,
+#endif
+#ifdef HAVE_KERNEL_SPLICE_READ
+        .splice_read    = ll_file_splice_read,
+#endif
         .fsync          = ll_fsync,
+        .flush          = ll_flush,
 #ifdef HAVE_F_OP_FLOCK
         .flock          = ll_file_flock,
 #endif
@@ -2519,13 +2589,23 @@ struct file_operations ll_file_operations_noflock = {
         .READ_METHOD    = READ_FUNCTION,
         .write          = ll_file_write,
         .WRITE_METHOD   = WRITE_FUNCTION,
+#ifdef HAVE_UNLOCKED_IOCTL
+        .unlocked_ioctl = ll_file_ioctl,
+#else
         .ioctl          = ll_file_ioctl,
+#endif
         .open           = ll_file_open,
         .release        = ll_file_release,
         .mmap           = ll_file_mmap,
         .llseek         = ll_file_seek,
+#ifdef HAVE_KERNEL_SENDFILE
         .sendfile       = ll_file_sendfile,
+#endif
+#ifdef HAVE_KERNEL_SPLICE_READ
+        .splice_read    = ll_file_splice_read,
+#endif
         .fsync          = ll_fsync,
+        .flush          = ll_flush,
 #ifdef HAVE_F_OP_FLOCK
         .flock          = ll_file_noflock,
 #endif
@@ -2533,9 +2613,6 @@ struct file_operations ll_file_operations_noflock = {
 };
 
 struct inode_operations ll_file_inode_operations = {
-#ifdef HAVE_VFS_INTENT_PATCHES
-        .setattr_raw    = ll_setattr_raw,
-#endif
         .setattr        = ll_setattr,
         .truncate       = ll_truncate,
         .getattr        = ll_getattr,
@@ -2544,12 +2621,15 @@ struct inode_operations ll_file_inode_operations = {
         .getxattr       = ll_getxattr,
         .listxattr      = ll_listxattr,
         .removexattr    = ll_removexattr,
+#ifdef  HAVE_LINUX_FIEMAP_H
+        .fiemap         = ll_fiemap,
+#endif
 };
 
 /* dynamic ioctl number support routins */
 static struct llioc_ctl_data {
-        struct rw_semaphore ioc_sem;
-        struct list_head    ioc_head;
+        cfs_rw_semaphore_t      ioc_sem;
+        cfs_list_t              ioc_head;
 } llioc = {
         __RWSEM_INITIALIZER(llioc.ioc_sem),
         CFS_LIST_HEAD_INIT(llioc.ioc_head)
@@ -2557,7 +2637,7 @@ static struct llioc_ctl_data {
 
 
 struct llioc_data {
-        struct list_head        iocd_list;
+        cfs_list_t              iocd_list;
         unsigned int            iocd_size;
         llioc_callback_t        iocd_cb;
         unsigned int            iocd_count;
@@ -2585,9 +2665,9 @@ void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
         in_data->iocd_count = count;
         memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
 
-        down_write(&llioc.ioc_sem);
-        list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
-        up_write(&llioc.ioc_sem);
+        cfs_down_write(&llioc.ioc_sem);
+        cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
+        cfs_up_write(&llioc.ioc_sem);
 
         RETURN(in_data);
 }
@@ -2599,19 +2679,19 @@ void ll_iocontrol_unregister(void *magic)
         if (magic == NULL)
                 return;
 
-        down_write(&llioc.ioc_sem);
-        list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
+        cfs_down_write(&llioc.ioc_sem);
+        cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
                 if (tmp == magic) {
                         unsigned int size = tmp->iocd_size;
 
-                        list_del(&tmp->iocd_list);
-                        up_write(&llioc.ioc_sem);
+                        cfs_list_del(&tmp->iocd_list);
+                        cfs_up_write(&llioc.ioc_sem);
 
                         OBD_FREE(tmp, size);
                         return;
                 }
         }
-        up_write(&llioc.ioc_sem);
+        cfs_up_write(&llioc.ioc_sem);
 
         CWARN("didn't find iocontrol register block with magic: %p\n", magic);
 }
@@ -2626,8 +2706,8 @@ enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file,
         struct llioc_data *data;
         int rc = -EINVAL, i;
 
-        down_read(&llioc.ioc_sem);
-        list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
+        cfs_down_read(&llioc.ioc_sem);
+        cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
                 for (i = 0; i < data->iocd_count; i++) {
                         if (cmd != data->iocd_cmd[i])
                                 continue;
@@ -2639,7 +2719,7 @@ enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file,
                 if (ret == LLIOC_STOP)
                         break;
         }
-        up_read(&llioc.ioc_sem);
+        cfs_up_read(&llioc.ioc_sem);
 
         if (rcp)
                 *rcp = rc;