Whamcloud - gitweb
LU-5396 llite: make some functions static
[fs/lustre-release.git] / lustre / llite / file.c
index dabc5d1..28cb292 100644 (file)
 
 #define DEBUG_SUBSYSTEM S_LLITE
 #include <lustre_dlm.h>
-#include <lustre_lite.h>
 #include <linux/pagemap.h>
 #include <linux/file.h>
+#include <linux/sched.h>
 #include "llite_internal.h"
 #include <lustre/ll_fiemap.h>
+#include <lustre_ioctl.h>
 
 #include "cl_object.h"
 
-struct ll_file_data *ll_file_data_get(void)
+static int
+ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
+
+static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
+                         bool *lease_broken);
+
+static enum llioc_iter
+ll_iocontrol_call(struct inode *inode, struct file *file,
+                 unsigned int cmd, unsigned long arg, int *rcp);
+
+static struct ll_file_data *ll_file_data_get(void)
 {
        struct ll_file_data *fd;
 
@@ -79,8 +90,7 @@ 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 =
-                                        ll_inode_to_ext_flags(inode->i_flags);
+       op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
         op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
         if (fh)
                 op_data->op_handle = *fh;
@@ -184,19 +194,10 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp,
                spin_unlock(&lli->lli_lock);
        }
 
-        if (rc == 0) {
-                rc = ll_objects_destroy(req, inode);
-                if (rc)
-                       CERROR("%s: inode "DFID
-                              " ll_objects destroy: rc = %d\n",
-                              ll_i2mdexp(inode)->exp_obd->obd_name,
-                              PFID(ll_inode2fid(inode)), rc);
-        }
-
        if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) {
                struct mdt_body *body;
                body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-               if (!(body->valid & OBD_MD_FLRELEASED))
+               if (!(body->mbo_valid & OBD_MD_FLRELEASED))
                        rc = -EBUSY;
        }
 
@@ -261,8 +262,8 @@ int ll_md_real_close(struct inode *inode, fmode_t fmode)
        RETURN(rc);
 }
 
-int ll_md_close(struct obd_export *md_exp, struct inode *inode,
-                struct file *file)
+static int ll_md_close(struct obd_export *md_exp, struct inode *inode,
+                      struct file *file)
 {
         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
         struct ll_inode_info *lli = ll_i2info(inode);
@@ -322,10 +323,11 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode,
                         rc = ll_md_real_close(file->f_dentry->d_inode,
                                               fd->fd_omode);
                 }
-        } else {
-                CERROR("Releasing a file %p with negative dentry %p. Name %s",
-                       file, file->f_dentry, file->f_dentry->d_name.name);
-        }
+       } else {
+               CERROR("released file has negative dentry: file = %p, "
+                      "dentry = %p, name = %s\n",
+                      file, file->f_dentry, file->f_dentry->d_name.name);
+       }
 
 out:
        LUSTRE_FPRIVATE(file) = NULL;
@@ -370,12 +372,10 @@ int ll_file_release(struct inode *inode, struct file *file)
         fd = LUSTRE_FPRIVATE(file);
         LASSERT(fd != NULL);
 
-        /* The last ref on @file, maybe not the the owner pid of statahead.
-         * Different processes can open the same dir, "ll_opendir_key" means:
-         * it is me that should stop the statahead thread. */
-        if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd &&
-            lli->lli_opendir_pid != 0)
-                ll_stop_statahead(inode, lli->lli_opendir_key);
+       /* The last ref on @file, maybe not the the owner pid of statahead,
+        * because parent and child process can share the same file handle. */
+       if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
+               ll_deauthorize_statahead(inode, fd);
 
         if (inode->i_sb->s_root == file->f_dentry) {
                 LUSTRE_FPRIVATE(file) = NULL;
@@ -384,7 +384,8 @@ int ll_file_release(struct inode *inode, struct file *file)
         }
 
         if (!S_ISDIR(inode->i_mode)) {
-               lov_read_and_clear_async_rc(lli->lli_clob);
+               if (lli->lli_clob != NULL)
+                       lov_read_and_clear_async_rc(lli->lli_clob);
                 lli->lli_async_rc = 0;
         }
 
@@ -396,77 +397,69 @@ int ll_file_release(struct inode *inode, struct file *file)
         RETURN(rc);
 }
 
-static int ll_intent_file_open(struct file *file, void *lmm,
-                               int lmmsize, struct lookup_intent *itp)
+static int ll_intent_file_open(struct file *file, void *lmm, int lmmsize,
+                               struct lookup_intent *itp)
 {
-        struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode);
-        struct dentry *parent = file->f_dentry->d_parent;
-        const char *name = file->f_dentry->d_name.name;
-        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;
+       struct dentry *de = file->f_dentry;
+       struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
+       struct dentry *parent = de->d_parent;
+       const char *name = NULL;
+       int len = 0;
+       struct md_op_data *op_data;
+       struct ptlrpc_request *req = NULL;
+       int rc;
+       ENTRY;
 
-        if (!parent)
-                RETURN(-ENOENT);
-
-        /* Usually we come here only for NFSD, and we want open lock.
-           But we can also get here with pre 2.6.15 patchless kernels, and in
-           that case that lock is also ok */
-        /* We can also get here if there was cached open handle in revalidate_it
-         * but it disappeared while we were getting from there to ll_file_open.
-         * But this means this file was closed and immediatelly opened which
-         * 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 == NULL && lmmsize == 0) {
-                itp->it_flags |= MDS_OPEN_LOCK;
-                if (itp->it_flags & FMODE_WRITE)
-                        opc = LUSTRE_OPC_CREATE;
-        }
+       LASSERT(parent != NULL);
+       LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
 
-        op_data  = ll_prep_md_op_data(NULL, parent->d_inode,
-                                      file->f_dentry->d_inode, name, len,
-                                      O_RDWR, opc, NULL);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
+       /* if server supports open-by-fid, or file name is invalid, don't pack
+        * name in open request */
+       if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID) &&
+           lu_name_is_valid_2(de->d_name.name, de->d_name.len)) {
+               name = de->d_name.name;
+               len = de->d_name.len;
+       }
 
-       itp->it_flags |= MDS_OPEN_BY_FID;
-        rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp,
-                            0 /*unused */, &req, ll_md_blocking_ast, 0);
-        ll_finish_md_op_data(op_data);
-        if (rc == -ESTALE) {
-                /* reason for keep own exit path - don`t flood log
-                * with messages with -ESTALE errors.
-                */
-                if (!it_disposition(itp, DISP_OPEN_OPEN) ||
-                     it_open_error(DISP_OPEN_OPEN, itp))
-                        GOTO(out, rc);
-                ll_release_openhandle(file->f_dentry, itp);
-                GOTO(out, rc);
-        }
+       op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
+                                    name, len, 0, LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
+       op_data->op_data = lmm;
+       op_data->op_data_size = lmmsize;
+
+       rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
+                           &ll_md_blocking_ast, 0);
+       ll_finish_md_op_data(op_data);
+       if (rc == -ESTALE) {
+               /* reason for keep own exit path - don`t flood log
+                * with messages with -ESTALE errors.
+                */
+               if (!it_disposition(itp, DISP_OPEN_OPEN) ||
+                    it_open_error(DISP_OPEN_OPEN, itp))
+                       GOTO(out, rc);
+               ll_release_openhandle(de, itp);
+               GOTO(out, rc);
+       }
 
-        if (it_disposition(itp, DISP_LOOKUP_NEG))
-                GOTO(out, rc = -ENOENT);
+       if (it_disposition(itp, DISP_LOOKUP_NEG))
+               GOTO(out, rc = -ENOENT);
 
-        if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
-                rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
-                CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
-                GOTO(out, rc);
-        }
+       if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
+               rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
+               CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
+               GOTO(out, rc);
+       }
 
-        rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL, itp);
-        if (!rc && itp->d.lustre.it_lock_mode)
-                ll_set_lock_data(sbi->ll_md_exp, file->f_dentry->d_inode,
-                                 itp, NULL);
+       rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
+       if (!rc && itp->d.lustre.it_lock_mode)
+               ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, NULL);
 
 out:
        ptlrpc_req_finished(req);
-        ll_intent_drop_lock(itp);
+       ll_intent_drop_lock(itp);
 
-        RETURN(rc);
+       RETURN(rc);
 }
 
 /**
@@ -490,8 +483,8 @@ static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
        struct mdt_body *body;
 
        body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-       och->och_fh = body->handle;
-       och->och_fid = body->fid1;
+       och->och_fh = body->mbo_handle;
+       och->och_fid = body->mbo_fid1;
        och->och_lease_handle.cookie = it->d.lustre.it_lock_handle;
        och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
        och->och_flags = it->it_flags;
@@ -499,8 +492,8 @@ static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
        return md_set_open_replay_data(md_exp, och, it);
 }
 
-int ll_local_open(struct file *file, struct lookup_intent *it,
-                  struct ll_file_data *fd, struct obd_client_handle *och)
+static int ll_local_open(struct file *file, struct lookup_intent *it,
+                        struct ll_file_data *fd, struct obd_client_handle *och)
 {
         struct inode *inode = file->f_dentry->d_inode;
         struct ll_inode_info *lli = ll_i2info(inode);
@@ -520,13 +513,17 @@ int ll_local_open(struct file *file, struct lookup_intent *it,
                        RETURN(rc);
 
                body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-               ll_ioepoch_open(lli, body->ioepoch);
+               ll_ioepoch_open(lli, body->mbo_ioepoch);
        }
 
        LUSTRE_FPRIVATE(file) = fd;
        ll_readahead_init(inode, &fd->fd_ras);
        fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
 
+       /* ll_cl_context initialize */
+       rwlock_init(&fd->fd_lock);
+       INIT_LIST_HEAD(&fd->fd_lccs);
+
        RETURN(0);
 }
 
@@ -545,36 +542,28 @@ int ll_local_open(struct file *file, struct lookup_intent *it,
  */
 int ll_file_open(struct inode *inode, struct file *file)
 {
-        struct ll_inode_info *lli = ll_i2info(inode);
-        struct lookup_intent *it, oit = { .it_op = IT_OPEN,
-                                          .it_flags = file->f_flags };
-        struct obd_client_handle **och_p = NULL;
-        __u64 *och_usecount = NULL;
-        struct ll_file_data *fd;
-        int rc = 0, opendir_set = 0;
-        ENTRY;
+       struct ll_inode_info *lli = ll_i2info(inode);
+       struct lookup_intent *it, oit = { .it_op = IT_OPEN,
+                                         .it_flags = file->f_flags };
+       struct obd_client_handle **och_p = NULL;
+       __u64 *och_usecount = NULL;
+       struct ll_file_data *fd;
+       int rc = 0;
+       ENTRY;
 
        CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
               PFID(ll_inode2fid(inode)), inode, file->f_flags);
 
-        it = file->private_data; /* XXX: compat macro */
-        file->private_data = NULL; /* prevent ll_local_open assertion */
+       it = file->private_data; /* XXX: compat macro */
+       file->private_data = NULL; /* prevent ll_local_open assertion */
 
        fd = ll_file_data_get();
        if (fd == NULL)
                GOTO(out_openerr, rc = -ENOMEM);
 
        fd->fd_file = file;
-       if (S_ISDIR(inode->i_mode)) {
-               spin_lock(&lli->lli_sa_lock);
-               if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL &&
-                   lli->lli_opendir_pid == 0) {
-                       lli->lli_opendir_key = fd;
-                       lli->lli_opendir_pid = current_pid();
-                       opendir_set = 1;
-               }
-               spin_unlock(&lli->lli_sa_lock);
-       }
+       if (S_ISDIR(inode->i_mode))
+               ll_authorize_statahead(inode, fd);
 
         if (inode->i_sb->s_root == file->f_dentry) {
                 LUSTRE_FPRIVATE(file) = fd;
@@ -653,9 +642,20 @@ restart:
                            would attempt to grab och_mutex as well, that would
                            result in a deadlock */
                        mutex_unlock(&lli->lli_och_mutex);
-                        it->it_create_mode |= M_CHECK_STALE;
+                       /*
+                        * Normally called under two situations:
+                        * 1. NFS export.
+                        * 2. A race/condition on MDS resulting in no open
+                        *    handle to be returned from LOOKUP|OPEN request,
+                        *    for example if the target entry was a symlink.
+                        *
+                        * Always fetch MDS_OPEN_LOCK if this is not setstripe.
+                        *
+                        * Always specify MDS_OPEN_BY_FID because we don't want
+                        * to get file with different fid.
+                        */
+                       it->it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID;
                         rc = ll_intent_file_open(file, NULL, 0, it);
-                        it->it_create_mode &= ~M_CHECK_STALE;
                         if (rc)
                                 GOTO(out_openerr, rc);
 
@@ -714,10 +714,10 @@ out_och_free:
                mutex_unlock(&lli->lli_och_mutex);
 
 out_openerr:
-                if (opendir_set != 0)
-                        ll_stop_statahead(inode, lli->lli_opendir_key);
-                if (fd != NULL)
-                        ll_file_data_put(fd);
+               if (lli->lli_opendir_key == fd)
+                       ll_deauthorize_statahead(inode, fd);
+               if (fd != NULL)
+                       ll_file_data_put(fd);
         } else {
                 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
         }
@@ -756,13 +756,14 @@ static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
 /**
  * Acquire a lease and open the file.
  */
-struct obd_client_handle *ll_lease_open(struct inode *inode, struct file *file,
-                                       fmode_t fmode, __u64 open_flags)
+static struct obd_client_handle *
+ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
+             __u64 open_flags)
 {
        struct lookup_intent it = { .it_op = IT_OPEN };
        struct ll_sb_info *sbi = ll_i2sbi(inode);
        struct md_op_data *op_data;
-       struct ptlrpc_request *req;
+       struct ptlrpc_request *req = NULL;
        struct lustre_handle old_handle = { 0 };
        struct obd_client_handle *och = NULL;
        int rc;
@@ -828,15 +829,15 @@ struct obd_client_handle *ll_lease_open(struct inode *inode, struct file *file,
 
        it.it_flags = fmode | open_flags;
        it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
-       rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, &req,
-                               ll_md_blocking_lease_ast,
+       rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
+                           &ll_md_blocking_lease_ast,
        /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
         * it can be cancelled which may mislead applications that the lease is
         * broken;
         * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
         * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
         * doesn't deal with openhandle, so normal openhandle will be leaked. */
-                               LDLM_FL_NO_LRU | LDLM_FL_EXCL);
+                           LDLM_FL_NO_LRU | LDLM_FL_EXCL);
        ll_finish_md_op_data(op_data);
        ptlrpc_req_finished(req);
        if (rc < 0)
@@ -860,7 +861,7 @@ struct obd_client_handle *ll_lease_open(struct inode *inode, struct file *file,
        if (it.d.lustre.it_lock_mode == 0 ||
            it.d.lustre.it_lock_bits != MDS_INODELOCK_OPEN) {
                /* open lock must return for lease */
-               CERROR(DFID "lease granted but no open lock, %d/%Lu.\n",
+               CERROR(DFID "lease granted but no open lock, %d/"LPU64".\n",
                        PFID(ll_inode2fid(inode)), it.d.lustre.it_lock_mode,
                        it.d.lustre.it_lock_bits);
                GOTO(out_close, rc = -EPROTO);
@@ -890,14 +891,13 @@ out:
                OBD_FREE_PTR(och);
        RETURN(ERR_PTR(rc));
 }
-EXPORT_SYMBOL(ll_lease_open);
 
 /**
  * Release lease and close the file.
  * It will check if the lease has ever broken.
  */
-int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
-                       bool *lease_broken)
+static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
+                         bool *lease_broken)
 {
        struct ldlm_lock *lock;
        bool cancelled = true;
@@ -924,7 +924,6 @@ int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
                                       NULL);
        RETURN(rc);
 }
-EXPORT_SYMBOL(ll_lease_close);
 
 /* Fills the obdo with the attributes for the lsm */
 static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
@@ -958,11 +957,11 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
                        oinfo.oi_oa->o_flags |= OBD_FL_FLUSH;
        }
 
-        set = ptlrpc_prep_set();
-        if (set == NULL) {
-                CERROR("can't allocate ptlrpc set\n");
-                rc = -ENOMEM;
-        } else {
+       set = ptlrpc_prep_set();
+       if (set == NULL) {
+               CERROR("cannot allocate ptlrpc set: rc = %d\n", -ENOMEM);
+               rc = -ENOMEM;
+       } else {
                 rc = obd_getattr_async(exp, &oinfo, set);
                 if (rc == 0)
                         rc = ptlrpc_set_wait(set);
@@ -1004,7 +1003,7 @@ int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
                CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu,"
                       " blksize %lu\n", POSTID(oi), i_size_read(inode),
                       (unsigned long long)inode->i_blocks,
-                      (unsigned long)ll_inode_blksize(inode));
+                      1UL << inode->i_blkbits);
        }
        ccc_inode_lsm_put(inode, lsm);
        RETURN(rc);
@@ -1026,7 +1025,12 @@ int ll_merge_lvb(const struct lu_env *env, struct inode *inode)
        LTIME_S(inode->i_atime) = lli->lli_lvb.lvb_atime;
        LTIME_S(inode->i_mtime) = lli->lli_lvb.lvb_mtime;
        LTIME_S(inode->i_ctime) = lli->lli_lvb.lvb_ctime;
-       inode_init_lvb(inode, &lvb);
+
+       lvb.lvb_size = i_size_read(inode);
+       lvb.lvb_blocks = inode->i_blocks;
+       lvb.lvb_mtime = LTIME_S(inode->i_mtime);
+       lvb.lvb_atime = LTIME_S(inode->i_atime);
+       lvb.lvb_ctime = LTIME_S(inode->i_ctime);
 
        cl_object_attr_lock(obj);
        rc = cl_object_attr_get(env, obj, attr);
@@ -1099,7 +1103,7 @@ static bool file_is_noatime(const struct file *file)
        return false;
 }
 
-void ll_io_init(struct cl_io *io, const struct file *file, int write)
+static void ll_io_init(struct cl_io *io, const struct file *file, int write)
 {
         struct inode *inode = file->f_dentry->d_inode;
 
@@ -1131,9 +1135,10 @@ ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
        struct ll_file_data  *fd  = LUSTRE_FPRIVATE(file);
        struct cl_io         *io;
        ssize_t               result;
+       struct range_lock     range;
        ENTRY;
 
-       CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: "LPU64", count: %zd\n",
+       CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: "LPU64", count: %zu\n",
                file->f_dentry->d_name.name, iot, *ppos, count);
 
 restart:
@@ -1141,12 +1146,16 @@ restart:
         ll_io_init(io, file, iot == CIT_WRITE);
 
         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);
-                int write_mutex_locked = 0;
+               struct vvp_io *vio = vvp_env_io(env);
+               struct ccc_io *cio = ccc_env_io(env);
+               bool range_locked = false;
 
-                cio->cui_fd  = LUSTRE_FPRIVATE(file);
-                vio->cui_io_subtype = args->via_io_subtype;
+               if (file->f_flags & O_APPEND)
+                       range_lock_init(&range, 0, LUSTRE_EOF);
+               else
+                       range_lock_init(&range, *ppos, *ppos + count - 1);
+               cio->cui_fd  = LUSTRE_FPRIVATE(file);
+               vio->cui_io_subtype = args->via_io_subtype;
 
                 switch (vio->cui_io_subtype) {
                 case IO_NORMAL:
@@ -1156,17 +1165,17 @@ restart:
                         cio->cui_iocb = args->u.normal.via_iocb;
                         if ((iot == CIT_WRITE) &&
                             !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
-                               if (mutex_lock_interruptible(&lli->
-                                                       lli_write_mutex))
-                                       GOTO(out, result = -ERESTARTSYS);
-                               write_mutex_locked = 1;
+                               CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
+                                      RL_PARA(&range));
+                               result = range_lock(&lli->lli_write_tree,
+                                                   &range);
+                               if (result < 0)
+                                       GOTO(out, result);
+
+                               range_locked = true;
                        }
                        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;
@@ -1175,11 +1184,18 @@ restart:
                         CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
                         LBUG();
                 }
+
+               ll_cl_add(file, env, io);
                 result = cl_io_loop(env, io);
+               ll_cl_remove(file, env);
+
                if (args->via_io_subtype == IO_NORMAL)
                        up_read(&lli->lli_trunc_sem);
-               if (write_mutex_locked)
-                       mutex_unlock(&lli->lli_write_mutex);
+               if (range_locked) {
+                       CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
+                              RL_PARA(&range));
+                       range_unlock(&lli->lli_write_tree, &range);
+               }
         } else {
                 /* cl_io_rw_init() handled IO */
                 result = io->ci_result;
@@ -1195,10 +1211,10 @@ out:
        /* If any bit been read/written (result != 0), we just return
         * short read/write instead of restart io. */
        if ((result == 0 || result == -ENODATA) && io->ci_need_restart) {
-               CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n",
+               CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zu\n",
                       iot == CIT_READ ? "read" : "write",
                       file->f_dentry->d_name.name, *ppos, count);
-               LASSERTF(io->ci_nob == 0, "%zd", io->ci_nob);
+               LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob);
                goto restart;
        }
 
@@ -1281,7 +1297,7 @@ static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
         RETURN(result);
 }
 
-static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
+static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
                             loff_t *ppos)
 {
         struct lu_env *env;
@@ -1347,8 +1363,8 @@ static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
         RETURN(result);
 }
 
-static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
-                             loff_t *ppos)
+static ssize_t ll_file_write(struct file *file, const char __user *buf,
+                            size_t count, loff_t *ppos)
 {
         struct lu_env *env;
         struct iovec  *local_iov;
@@ -1406,88 +1422,6 @@ static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
         RETURN(result);
 }
 
-static int ll_lov_recreate(struct inode *inode, struct ost_id *oi,
-                           obd_count ost_idx)
-{
-       struct obd_export *exp = ll_i2dtexp(inode);
-       struct obd_trans_info oti = { 0 };
-       struct obdo *oa = NULL;
-       int lsm_size;
-       int rc = 0;
-       struct lov_stripe_md *lsm = NULL, *lsm2;
-       ENTRY;
-
-       OBDO_ALLOC(oa);
-       if (oa == NULL)
-               RETURN(-ENOMEM);
-
-       lsm = ccc_inode_lsm_get(inode);
-       if (!lsm_has_objects(lsm))
-                GOTO(out, rc = -ENOENT);
-
-        lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
-                   (lsm->lsm_stripe_count));
-
-        OBD_ALLOC_LARGE(lsm2, lsm_size);
-        if (lsm2 == NULL)
-                GOTO(out, rc = -ENOMEM);
-
-       oa->o_oi = *oi;
-        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_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
-        memcpy(lsm2, lsm, lsm_size);
-       ll_inode_size_lock(inode);
-       rc = obd_create(NULL, exp, oa, &lsm2, &oti);
-       ll_inode_size_unlock(inode);
-
-       OBD_FREE_LARGE(lsm2, lsm_size);
-       GOTO(out, rc);
-out:
-       ccc_inode_lsm_put(inode, lsm);
-       OBDO_FREE(oa);
-       return rc;
-}
-
-static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
-{
-       struct ll_recreate_obj ucreat;
-       struct ost_id           oi;
-       ENTRY;
-
-       if (!cfs_capable(CFS_CAP_SYS_ADMIN))
-               RETURN(-EPERM);
-
-       if (copy_from_user(&ucreat, (struct ll_recreate_obj *)arg,
-                          sizeof(ucreat)))
-               RETURN(-EFAULT);
-
-       ostid_set_seq_mdt0(&oi);
-       ostid_set_id(&oi, ucreat.lrc_id);
-       RETURN(ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx));
-}
-
-static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
-{
-       struct lu_fid   fid;
-       struct ost_id   oi;
-       obd_count       ost_idx;
-        ENTRY;
-
-       if (!cfs_capable(CFS_CAP_SYS_ADMIN))
-               RETURN(-EPERM);
-
-       if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid)))
-               RETURN(-EFAULT);
-
-       fid_to_ostid(&fid, &oi);
-       ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
-       RETURN(ll_lov_recreate(inode, &oi, ost_idx));
-}
-
 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
                              __u64  flags, struct lov_user_md *lum,
                             int lum_size)
@@ -1506,6 +1440,7 @@ int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
        }
 
        ll_inode_size_lock(inode);
+       oit.it_flags |= MDS_OPEN_BY_FID;
        rc = ll_intent_file_open(file, lum, lum_size, &oit);
        if (rc)
                GOTO(out_unlock, rc);
@@ -1538,9 +1473,9 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
         struct md_op_data *op_data;
         int rc, lmmsize;
 
-        rc = ll_get_max_mdsize(sbi, &lmmsize);
-        if (rc)
-                RETURN(rc);
+       rc = ll_get_default_mdsize(sbi, &lmmsize);
+       if (rc)
+               RETURN(rc);
 
         op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
                                      strlen(filename), lmmsize,
@@ -1560,9 +1495,9 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
         LASSERT(body != NULL); /* checked by mdc_getattr_name */
 
-        lmmsize = body->eadatasize;
+       lmmsize = body->mbo_eadatasize;
 
-        if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
+       if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
                         lmmsize == 0) {
                 GOTO(out, rc = -ENODATA);
         }
@@ -1591,13 +1526,14 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
                  * avoid swab not existent lsm objects */
                 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
-                        if (S_ISREG(body->mode))
-                                lustre_swab_lov_user_md_objects(
-                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
-                                 stripe_count);
-                } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
-                        lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
-                        if (S_ISREG(body->mode))
+                       if (S_ISREG(body->mbo_mode))
+                               lustre_swab_lov_user_md_objects(
+                                   ((struct lov_user_md_v1 *)lmm)->lmm_objects,
+                                   stripe_count);
+               } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
+                       lustre_swab_lov_user_md_v3(
+                               (struct lov_user_md_v3 *)lmm);
+                       if (S_ISREG(body->mbo_mode))
                                 lustre_swab_lov_user_md_objects(
                                  ((struct lov_user_md_v3 *)lmm)->lmm_objects,
                                  stripe_count);
@@ -1628,7 +1564,7 @@ static int ll_lov_setea(struct inode *inode, struct file *file,
        if (lump == NULL)
                 RETURN(-ENOMEM);
 
-       if (copy_from_user(lump, (struct lov_user_md  *)arg, lum_size)) {
+       if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) {
                OBD_FREE_LARGE(lump, lum_size);
                RETURN(-EFAULT);
        }
@@ -1642,38 +1578,32 @@ static int ll_lov_setea(struct inode *inode, struct file *file,
 static int ll_lov_setstripe(struct inode *inode, struct file *file,
                            unsigned long arg)
 {
-       struct lov_user_md_v3    lumv3;
-       struct lov_user_md_v1   *lumv1 = (struct lov_user_md_v1 *)&lumv3;
-       struct lov_user_md_v1   *lumv1p = (struct lov_user_md_v1 *)arg;
-       struct lov_user_md_v3   *lumv3p = (struct lov_user_md_v3 *)arg;
-       int                      lum_size, rc;
-       __u64                    flags = FMODE_WRITE;
+       struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
+       struct lov_user_md        *klum;
+       int                        lum_size, rc;
+       __u64                      flags = FMODE_WRITE;
        ENTRY;
 
-       /* 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))
-               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))
-                       RETURN(-EFAULT);
-       }
+       rc = ll_copy_user_md(lum, &klum);
+       if (rc < 0)
+               RETURN(rc);
 
-       rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size);
+       lum_size = rc;
+       rc = ll_lov_setstripe_ea_info(inode, file, flags, klum, lum_size);
        if (rc == 0) {
                struct lov_stripe_md *lsm;
                __u32 gen;
 
-               put_user(0, &lumv1p->lmm_stripe_count);
+               put_user(0, &lum->lmm_stripe_count);
 
                ll_layout_refresh(inode, &gen);
                lsm = ccc_inode_lsm_get(inode);
                rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
-                                  0, lsm, (void *)arg);
+                                  0, lsm, lum);
                ccc_inode_lsm_put(inode, lsm);
        }
+
+       OBD_FREE(klum, lum_size);
        RETURN(rc);
 }
 
@@ -1686,12 +1616,13 @@ static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
        lsm = ccc_inode_lsm_get(inode);
        if (lsm != NULL)
                rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
-                                  lsm, (void *)arg);
+                                  lsm, (void __user *)arg);
        ccc_inode_lsm_put(inode, lsm);
        RETURN(rc);
 }
 
-int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
+static int
+ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
 {
         struct ll_inode_info   *lli = ll_i2info(inode);
         struct ll_file_data    *fd = LUSTRE_FPRIVATE(file);
@@ -1699,6 +1630,11 @@ int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
         int                     rc;
         ENTRY;
 
+       if (arg == 0) {
+               CWARN("group id for group lock must not be 0\n");
+               RETURN(-EINVAL);
+       }
+
         if (ll_file_nolock(file))
                 RETURN(-EOPNOTSUPP);
 
@@ -1733,7 +1669,8 @@ int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
        RETURN(0);
 }
 
-int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg)
+static int ll_put_grouplock(struct inode *inode, struct file *file,
+                           unsigned long arg)
 {
        struct ll_inode_info   *lli = ll_i2info(inode);
        struct ll_file_data    *fd = LUSTRE_FPRIVATE(file);
@@ -1814,13 +1751,13 @@ out:
  * Get size for inode for which FIEMAP mapping is requested.
  * Make the FIEMAP get_info call and returns the result.
  */
-int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
-              int num_bytes)
+static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
+                       size_t num_bytes)
 {
        struct obd_export *exp = ll_i2dtexp(inode);
        struct lov_stripe_md *lsm = NULL;
         struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
-        int vallen = num_bytes;
+       __u32 vallen = num_bytes;
         int rc;
         ENTRY;
 
@@ -1871,38 +1808,39 @@ out:
        RETURN(rc);
 }
 
-int ll_fid2path(struct inode *inode, void *arg)
+int ll_fid2path(struct inode *inode, void __user *arg)
 {
        struct obd_export       *exp = ll_i2mdexp(inode);
-       struct getinfo_fid2path *gfout, *gfin;
-       int                      outsize, rc;
+       const struct getinfo_fid2path __user *gfin = arg;
+       __u32                    pathlen;
+       struct getinfo_fid2path *gfout;
+       size_t                   outsize;
+       int                      rc;
+
        ENTRY;
 
        if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
            !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
                RETURN(-EPERM);
 
-       /* Need to get the buflen */
-       OBD_ALLOC_PTR(gfin);
-       if (gfin == NULL)
-               RETURN(-ENOMEM);
-       if (copy_from_user(gfin, arg, sizeof(*gfin))) {
-               OBD_FREE_PTR(gfin);
+       /* Only need to get the buflen */
+       if (get_user(pathlen, &gfin->gf_pathlen))
                RETURN(-EFAULT);
-       }
 
-       outsize = sizeof(*gfout) + gfin->gf_pathlen;
+       if (pathlen > PATH_MAX)
+               RETURN(-EINVAL);
+
+       outsize = sizeof(*gfout) + pathlen;
        OBD_ALLOC(gfout, outsize);
-       if (gfout == NULL) {
-               OBD_FREE_PTR(gfin);
+       if (gfout == NULL)
                RETURN(-ENOMEM);
-       }
-       memcpy(gfout, gfin, sizeof(*gfout));
-       OBD_FREE_PTR(gfin);
+
+       if (copy_from_user(gfout, arg, sizeof(*gfout)))
+               GOTO(gf_free, rc = -EFAULT);
 
        /* Call mdc_iocontrol */
        rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
-       if (rc)
+       if (rc != 0)
                GOTO(gf_free, rc);
 
        if (copy_to_user(arg, gfout, outsize))
@@ -1925,6 +1863,10 @@ static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
         if (get_user(extent_count,
             &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
                 RETURN(-EFAULT);
+
+       if (extent_count >=
+           (SIZE_MAX - sizeof(*fiemap_s)) / sizeof(struct ll_fiemap_extent))
+               RETURN(-EINVAL);
         num_bytes = sizeof(*fiemap_s) + (extent_count *
                                          sizeof(struct ll_fiemap_extent));
 
@@ -1957,7 +1899,7 @@ static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
                 ret_bytes += (fiemap_s->fm_mapped_extents *
                                  sizeof(struct ll_fiemap_extent));
 
-       if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
+       if (copy_to_user((void __user *)arg, fiemap_s, ret_bytes))
                rc = -EFAULT;
 
 error:
@@ -2272,8 +2214,8 @@ static int ll_hsm_import(struct inode *inode, struct file *file,
 
        attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
        attr->ia_mode |= S_IFREG;
-       attr->ia_uid = hui->hui_uid;
-       attr->ia_gid = hui->hui_gid;
+       attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
+       attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
        attr->ia_size = hui->hui_size;
        attr->ia_mtime.tv_sec = hui->hui_mtime;
        attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
@@ -2285,10 +2227,14 @@ static int ll_hsm_import(struct inode *inode, struct file *file,
                         ATTR_MTIME | ATTR_MTIME_SET |
                         ATTR_ATIME | ATTR_ATIME_SET;
 
+       mutex_lock(&inode->i_mutex);
+
        rc = ll_setattr_raw(file->f_dentry, attr, true);
        if (rc == -ENODATA)
                rc = 0;
 
+       mutex_unlock(&inode->i_mutex);
+
 out:
        if (hss != NULL)
                OBD_FREE_PTR(hss);
@@ -2299,7 +2245,14 @@ out:
        RETURN(rc);
 }
 
-long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+static inline long ll_lease_type_from_fmode(fmode_t fmode)
+{
+       return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
+              ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
+}
+
+static long
+ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 {
        struct inode            *inode = file->f_dentry->d_inode;
        struct ll_file_data     *fd = LUSTRE_FPRIVATE(file);
@@ -2317,14 +2270,14 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
         switch(cmd) {
         case LL_IOC_GETFLAGS:
                 /* Get the current value of the file flags */
-                return put_user(fd->fd_flags, (int *)arg);
+               return put_user(fd->fd_flags, (int __user *)arg);
         case LL_IOC_SETFLAGS:
         case LL_IOC_CLRFLAGS:
                 /* Set or clear specific file flags */
                 /* XXX This probably needs checks to ensure the flags are
                  *     not abused, and to handle any flag side effects.
                  */
-                if (get_user(flags, (int *) arg))
+               if (get_user(flags, (int __user *) arg))
                         RETURN(-EFAULT);
 
                 if (cmd == LL_IOC_SETFLAGS) {
@@ -2348,7 +2301,7 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                struct file *file2;
                struct lustre_swap_layouts lsl;
 
-               if (copy_from_user(&lsl, (char *)arg,
+               if (copy_from_user(&lsl, (char __user *)arg,
                                       sizeof(struct lustre_swap_layouts)))
                        RETURN(-EFAULT);
 
@@ -2367,10 +2320,6 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
        }
         case LL_IOC_LOV_GETSTRIPE:
                 RETURN(ll_lov_getstripe(inode, arg));
-        case LL_IOC_RECREATE_OBJ:
-                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:
@@ -2378,13 +2327,13 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                 RETURN(ll_iocontrol(inode, file, cmd, arg));
         case FSFILT_IOC_GETVERSION_OLD:
         case FSFILT_IOC_GETVERSION:
-                RETURN(put_user(inode->i_generation, (int *)arg));
+               RETURN(put_user(inode->i_generation, (int __user *)arg));
         case LL_IOC_GROUP_LOCK:
                 RETURN(ll_get_grouplock(inode, file, arg));
         case LL_IOC_GROUP_UNLOCK:
                 RETURN(ll_put_grouplock(inode, file, arg));
         case IOC_OBD_STATFS:
-                RETURN(ll_obd_statfs(inode, (void *)arg));
+               RETURN(ll_obd_statfs(inode, (void __user *)arg));
 
         /* We need to special case any other ioctls we want to handle,
          * to send them to the MDS/OST as appropriate and to properly
@@ -2395,25 +2344,29 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
        case LL_IOC_FLUSHCTX:
                RETURN(ll_flush_ctx(inode));
        case LL_IOC_PATH2FID: {
-               if (copy_to_user((void *)arg, ll_inode2fid(inode),
+               if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
                                 sizeof(struct lu_fid)))
                        RETURN(-EFAULT);
 
                RETURN(0);
        }
+       case LL_IOC_GETPARENT:
+               RETURN(ll_getparent(file, (void __user *)arg));
+
        case OBD_IOC_FID2PATH:
-               RETURN(ll_fid2path(inode, (void *)arg));
+               RETURN(ll_fid2path(inode, (void __user *)arg));
        case LL_IOC_DATA_VERSION: {
                struct ioc_data_version idv;
                int rc;
 
-               if (copy_from_user(&idv, (char *)arg, sizeof(idv)))
+               if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
                        RETURN(-EFAULT);
 
                idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
                rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
 
-               if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv)))
+               if (rc == 0 &&
+                   copy_to_user((char __user *)arg, &idv, sizeof(idv)))
                        RETURN(-EFAULT);
 
                RETURN(rc);
@@ -2426,7 +2379,7 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                 if (mdtidx < 0)
                         RETURN(mdtidx);
 
-                if (put_user((int)mdtidx, (int*)arg))
+               if (put_user((int)mdtidx, (int __user *)arg))
                         RETURN(-EFAULT);
 
                 RETURN(0);
@@ -2453,7 +2406,7 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
                                   op_data, NULL);
 
-               if (copy_to_user((void *)arg, hus, sizeof(*hus)))
+               if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
                        rc = -EFAULT;
 
                ll_finish_md_op_data(op_data);
@@ -2468,7 +2421,7 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                if (hss == NULL)
                        RETURN(-ENOMEM);
 
-               if (copy_from_user(hss, (char *)arg, sizeof(*hss))) {
+               if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
                        OBD_FREE_PTR(hss);
                        RETURN(-EFAULT);
                }
@@ -2497,7 +2450,7 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
                                   op_data, NULL);
 
-               if (copy_to_user((char *)arg, hca, sizeof(*hca)))
+               if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
                        rc = -EFAULT;
 
                ll_finish_md_op_data(op_data);
@@ -2508,20 +2461,20 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                struct ll_inode_info *lli = ll_i2info(inode);
                struct obd_client_handle *och = NULL;
                bool lease_broken;
-               fmode_t mode = 0;
+               fmode_t fmode;
 
                switch (arg) {
-               case F_WRLCK:
+               case LL_LEASE_WRLCK:
                        if (!(file->f_mode & FMODE_WRITE))
                                RETURN(-EPERM);
-                       mode = FMODE_WRITE;
+                       fmode = FMODE_WRITE;
                        break;
-               case F_RDLCK:
+               case LL_LEASE_RDLCK:
                        if (!(file->f_mode & FMODE_READ))
                                RETURN(-EPERM);
-                       mode = FMODE_READ;
+                       fmode = FMODE_READ;
                        break;
-               case F_UNLCK:
+               case LL_LEASE_UNLCK:
                        mutex_lock(&lli->lli_och_mutex);
                        if (fd->fd_lease_och != NULL) {
                                och = fd->fd_lease_och;
@@ -2529,25 +2482,26 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                        }
                        mutex_unlock(&lli->lli_och_mutex);
 
-                       if (och != NULL) {
-                               mode = och->och_flags &(FMODE_READ|FMODE_WRITE);
-                               rc = ll_lease_close(och, inode, &lease_broken);
-                               if (rc == 0 && lease_broken)
-                                       mode = 0;
-                       } else {
-                               rc = -ENOLCK;
-                       }
+                       if (och == NULL)
+                               RETURN(-ENOLCK);
+
+                       fmode = och->och_flags;
+                       rc = ll_lease_close(och, inode, &lease_broken);
+                       if (rc < 0)
+                               RETURN(rc);
 
-                       /* return the type of lease or error */
-                       RETURN(rc < 0 ? rc : (int)mode);
+                       if (lease_broken)
+                               fmode = 0;
+
+                       RETURN(ll_lease_type_from_fmode(fmode));
                default:
                        RETURN(-EINVAL);
                }
 
-               CDEBUG(D_INODE, "Set lease with mode %d\n", mode);
+               CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
 
                /* apply for lease */
-               och = ll_lease_open(inode, file, mode, 0);
+               och = ll_lease_open(inode, file, fmode, 0);
                if (IS_ERR(och))
                        RETURN(PTR_ERR(och));
 
@@ -2568,8 +2522,8 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
        case LL_IOC_GET_LEASE: {
                struct ll_inode_info *lli = ll_i2info(inode);
                struct ldlm_lock *lock = NULL;
+               fmode_t fmode = 0;
 
-               rc = 0;
                mutex_lock(&lli->lli_och_mutex);
                if (fd->fd_lease_och != NULL) {
                        struct obd_client_handle *och = fd->fd_lease_och;
@@ -2578,14 +2532,15 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                        if (lock != NULL) {
                                lock_res_and_lock(lock);
                                if (!ldlm_is_cancel(lock))
-                                       rc = och->och_flags &
-                                               (FMODE_READ | FMODE_WRITE);
+                                       fmode = och->och_flags;
+
                                unlock_res_and_lock(lock);
                                LDLM_LOCK_PUT(lock);
                        }
                }
                mutex_unlock(&lli->lli_och_mutex);
-               RETURN(rc);
+
+               RETURN(ll_lease_type_from_fmode(fmode));
        }
        case LL_IOC_HSM_IMPORT: {
                struct hsm_user_import *hui;
@@ -2594,7 +2549,7 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                if (hui == NULL)
                        RETURN(-ENOMEM);
 
-               if (copy_from_user(hui, (void *)arg, sizeof(*hui))) {
+               if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
                        OBD_FREE_PTR(hui);
                        RETURN(-EFAULT);
                }
@@ -2604,6 +2559,7 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                OBD_FREE_PTR(hui);
                RETURN(rc);
        }
+
        default: {
                int err;
 
@@ -2612,7 +2568,7 @@ long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                        RETURN(err);
 
                RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
-                                    (void *)arg));
+                                    (void __user *)arg));
        }
        }
 }
@@ -2684,7 +2640,7 @@ generic_file_llseek_size(struct file *file, loff_t offset, int origin,
 }
 #endif
 
-loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
+static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
 {
        struct inode *inode = file->f_dentry->d_inode;
        loff_t retval, eof = 0;
@@ -2709,7 +2665,7 @@ loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
        RETURN(retval);
 }
 
-int ll_flush(struct file *file, fl_owner_t id)
+static int ll_flush(struct file *file, fl_owner_t id)
 {
        struct inode *inode = file->f_dentry->d_inode;
        struct ll_inode_info *lli = ll_i2info(inode);
@@ -2722,9 +2678,11 @@ int ll_flush(struct file *file, fl_owner_t id)
         * failed for pages in this mapping. */
        rc = lli->lli_async_rc;
        lli->lli_async_rc = 0;
-       err = lov_read_and_clear_async_rc(lli->lli_clob);
-       if (rc == 0)
-               rc = err;
+       if (lli->lli_clob != NULL) {
+               err = lov_read_and_clear_async_rc(lli->lli_clob);
+               if (rc == 0)
+                       rc = err;
+       }
 
        /* The application has been told write failure already.
         * Do not report failure again. */
@@ -2735,7 +2693,7 @@ int ll_flush(struct file *file, fl_owner_t id)
 
 /**
  * Called to make sure a portion of file has been written out.
- * if @local_only is not true, it will send OST_SYNC RPCs to ost.
+ * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
  *
  * Return how many pages have been written.
  */
@@ -2801,56 +2759,59 @@ int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 int ll_fsync(struct file *file, int datasync)
 {
        struct dentry *dentry = file->f_dentry;
+       loff_t start = 0;
+       loff_t end = LLONG_MAX;
 #else
 int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
 {
+       loff_t start = 0;
+       loff_t end = LLONG_MAX;
 #endif
-        struct inode *inode = dentry->d_inode;
-        struct ll_inode_info *lli = ll_i2info(inode);
-        struct ptlrpc_request *req;
-        struct obd_capa *oc;
-        int rc, err;
-        ENTRY;
+       struct inode *inode = dentry->d_inode;
+       struct ll_inode_info *lli = ll_i2info(inode);
+       struct ptlrpc_request *req;
+       struct obd_capa *oc;
+       int rc, err;
+       ENTRY;
 
        CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
               PFID(ll_inode2fid(inode)), inode);
-        ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
+       ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
 
 #ifdef HAVE_FILE_FSYNC_4ARGS
        rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
        mutex_lock(&inode->i_mutex);
 #else
-        /* fsync's caller has already called _fdata{sync,write}, we want
-         * that IO to finish before calling the osc and mdc sync methods */
-        rc = filemap_fdatawait(inode->i_mapping);
+       /* fsync's caller has already called _fdata{sync,write}, we want
+        * that IO to finish before calling the osc and mdc sync methods */
+       rc = filemap_fdatawait(inode->i_mapping);
 #endif
 
-        /* catch async errors that were recorded back when async writeback
-         * failed for pages in this mapping. */
-        if (!S_ISDIR(inode->i_mode)) {
-                err = lli->lli_async_rc;
-                lli->lli_async_rc = 0;
-                if (rc == 0)
-                        rc = err;
+       /* catch async errors that were recorded back when async writeback
+        * failed for pages in this mapping. */
+       if (!S_ISDIR(inode->i_mode)) {
+               err = lli->lli_async_rc;
+               lli->lli_async_rc = 0;
+               if (rc == 0)
+                       rc = err;
                err = lov_read_and_clear_async_rc(lli->lli_clob);
                if (rc == 0)
                        rc = err;
-        }
+       }
 
-        oc = ll_mdscapa_get(inode);
+       oc = ll_mdscapa_get(inode);
        err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
                       &req);
-        capa_put(oc);
-        if (!rc)
-                rc = err;
-        if (!err)
-                ptlrpc_req_finished(req);
+       capa_put(oc);
+       if (!rc)
+               rc = err;
+       if (!err)
+               ptlrpc_req_finished(req);
 
-       if (datasync && S_ISREG(inode->i_mode)) {
+       if (S_ISREG(inode->i_mode)) {
                struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
 
-               err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
-                               CL_FSYNC_ALL, 0);
+               err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
                if (rc == 0 && err < 0)
                        rc = err;
                if (rc < 0)
@@ -2865,7 +2826,8 @@ int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
        RETURN(rc);
 }
 
-int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
+static int
+ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 {
        struct inode *inode = file->f_dentry->d_inode;
        struct ll_sb_info *sbi = ll_i2sbi(inode);
@@ -2877,6 +2839,7 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
        struct md_op_data *op_data;
        struct lustre_handle lockh = {0};
        ldlm_policy_data_t flock = {{0}};
+       int fl_type = file_lock->fl_type;
        __u64 flags = 0;
        int rc;
        int rc2 = 0;
@@ -2912,7 +2875,7 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
        if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
                flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
 
-        switch (file_lock->fl_type) {
+       switch (fl_type) {
         case F_RDLCK:
                 einfo.ei_mode = LCK_PR;
                 break;
@@ -2931,8 +2894,7 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
                 einfo.ei_mode = LCK_PW;
                 break;
         default:
-                CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n",
-                        file_lock->fl_type);
+               CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
                 RETURN (-ENOTSUPP);
         }
 
@@ -2954,15 +2916,16 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
         case F_GETLK64:
 #endif
                 flags = LDLM_FL_TEST_LOCK;
-                /* Save the old mode so that if the mode in the lock changes we
-                 * can decrement the appropriate reader or writer refcount. */
-                file_lock->fl_type = einfo.ei_mode;
                 break;
         default:
                 CERROR("unknown fcntl lock command: %d\n", cmd);
                 RETURN (-EINVAL);
         }
 
+       /* Save the old mode so that if the mode in the lock changes we
+        * can decrement the appropriate reader or writer refcount. */
+       file_lock->fl_type = einfo.ei_mode;
+
         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
                                      LUSTRE_OPC_ANY, NULL);
         if (IS_ERR(op_data))
@@ -2973,8 +2936,12 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
               flock.l_flock.pid, flags, einfo.ei_mode,
               flock.l_flock.start, flock.l_flock.end);
 
-        rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL,
-                        op_data, &lockh, &flock, 0, NULL /* req */, flags);
+       rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
+                       flags);
+
+       /* Restore the file lock type if not TEST lock. */
+       if (!(flags & LDLM_FL_TEST_LOCK))
+               file_lock->fl_type = fl_type;
 
         if ((file_lock->fl_flags & FL_FLOCK) &&
             (rc == 0 || file_lock->fl_type == F_UNLCK))
@@ -2986,8 +2953,8 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 
        if (rc2 && file_lock->fl_type != F_UNLCK) {
                einfo.ei_mode = LCK_NL;
-               md_enqueue(sbi->ll_md_exp, &einfo, NULL,
-                       op_data, &lockh, &flock, 0, NULL /* req */, flags);
+               md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
+                          &lockh, flags);
                rc = rc2;
        }
 
@@ -2996,7 +2963,114 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
         RETURN(rc);
 }
 
-int ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
+int ll_get_fid_by_name(struct inode *parent, const char *name,
+                      int namelen, struct lu_fid *fid)
+{
+       struct md_op_data       *op_data = NULL;
+       struct mdt_body         *body;
+       struct ptlrpc_request   *req;
+       int                     rc;
+       ENTRY;
+
+       op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
+                                    LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
+
+       op_data->op_valid = OBD_MD_FLID;
+       rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
+       ll_finish_md_op_data(op_data);
+       if (rc < 0)
+               RETURN(rc);
+
+       body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+       if (body == NULL)
+               GOTO(out_req, rc = -EFAULT);
+       if (fid != NULL)
+               *fid = body->mbo_fid1;
+out_req:
+       ptlrpc_req_finished(req);
+       RETURN(rc);
+}
+
+int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
+              const char *name, int namelen)
+{
+       struct dentry         *dchild = NULL;
+       struct inode          *child_inode = NULL;
+       struct md_op_data     *op_data;
+       struct ptlrpc_request *request = NULL;
+       struct qstr           qstr;
+       int                    rc;
+       ENTRY;
+
+       CDEBUG(D_VFSTRACE, "migrate %s under "DFID" to MDT%04x\n",
+              name, PFID(ll_inode2fid(parent)), mdtidx);
+
+       op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
+                                    0, LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
+
+       /* Get child FID first */
+       qstr.hash = full_name_hash(name, namelen);
+       qstr.name = name;
+       qstr.len = namelen;
+       dchild = d_lookup(file->f_dentry, &qstr);
+       if (dchild != NULL && dchild->d_inode != NULL) {
+               op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
+               if (dchild->d_inode != NULL) {
+                       child_inode = igrab(dchild->d_inode);
+                       ll_invalidate_aliases(child_inode);
+               }
+               dput(dchild);
+       } else {
+               rc = ll_get_fid_by_name(parent, name, namelen,
+                                       &op_data->op_fid3);
+               if (rc != 0)
+                       GOTO(out_free, rc);
+       }
+
+       if (!fid_is_sane(&op_data->op_fid3)) {
+               CERROR("%s: migrate %s , but fid "DFID" is insane\n",
+                      ll_get_fsname(parent->i_sb, NULL, 0), name,
+                      PFID(&op_data->op_fid3));
+               GOTO(out_free, rc = -EINVAL);
+       }
+
+       rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
+       if (rc < 0)
+               GOTO(out_free, rc);
+
+       if (rc == mdtidx) {
+               CDEBUG(D_INFO, "%s:"DFID" is already on MDT%d.\n", name,
+                      PFID(&op_data->op_fid3), mdtidx);
+               GOTO(out_free, rc = 0);
+       }
+
+       op_data->op_mds = mdtidx;
+       op_data->op_cli_flags = CLI_MIGRATE;
+       rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
+                      namelen, name, namelen, &request);
+       if (rc == 0)
+               ll_update_times(request, parent);
+
+       ptlrpc_req_finished(request);
+       if (rc != 0)
+               GOTO(out_free, rc);
+
+out_free:
+       if (child_inode != NULL) {
+               clear_nlink(child_inode);
+               iput(child_inode);
+       }
+
+       ll_finish_md_op_data(op_data);
+       RETURN(rc);
+}
+
+static int
+ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
 {
         ENTRY;
 
@@ -3083,16 +3157,16 @@ static int ll_inode_revalidate_fini(struct inode *inode, int rc)
                if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
                        return 0;
        } else if (rc != 0) {
-               CERROR("%s: revalidate FID "DFID" error: rc = %d\n",
-                      ll_get_fsname(inode->i_sb, NULL, 0),
-                      PFID(ll_inode2fid(inode)), rc);
+               CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
+                            "%s: revalidate FID "DFID" error: rc = %d\n",
+                            ll_get_fsname(inode->i_sb, NULL, 0),
+                            PFID(ll_inode2fid(inode)), rc);
        }
 
        return rc;
 }
 
-int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
-                             __u64 ibits)
+static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
 {
         struct inode *inode = dentry->d_inode;
         struct ptlrpc_request *req = NULL;
@@ -3124,14 +3198,9 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
                 if (IS_ERR(op_data))
                         RETURN(PTR_ERR(op_data));
 
-                oit.it_create_mode |= M_CHECK_STALE;
-                rc = md_intent_lock(exp, op_data, NULL, 0,
-                                    /* we are not interested in name
-                                       based lookup */
-                                    &oit, 0, &req,
-                                    ll_md_blocking_ast, 0);
+               rc = md_intent_lock(exp, op_data, &oit, &req,
+                                   &ll_md_blocking_ast, 0);
                 ll_finish_md_op_data(op_data);
-                oit.it_create_mode &= ~M_CHECK_STALE;
                 if (rc < 0) {
                         rc = ll_inode_revalidate_fini(inode, rc);
                         GOTO (out, rc);
@@ -3157,12 +3226,12 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
                 struct md_op_data *op_data;
                 int ealen = 0;
 
-                if (S_ISREG(inode->i_mode)) {
-                        rc = ll_get_max_mdsize(sbi, &ealen);
-                        if (rc)
-                                RETURN(rc);
-                        valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
-                }
+               if (S_ISREG(inode->i_mode)) {
+                       rc = ll_get_default_mdsize(sbi, &ealen);
+                       if (rc)
+                               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,
@@ -3209,14 +3278,14 @@ static int ll_merge_md_attr(struct inode *inode)
        RETURN(0);
 }
 
-int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
-                          __u64 ibits)
+static int
+ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
 {
        struct inode    *inode = dentry->d_inode;
        int              rc;
        ENTRY;
 
-       rc = __ll_inode_revalidate_it(dentry, it, ibits);
+       rc = __ll_inode_revalidate(dentry, ibits);
        if (rc != 0)
                RETURN(rc);
 
@@ -3246,16 +3315,15 @@ int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
        RETURN(rc);
 }
 
-int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
-                  struct lookup_intent *it, struct kstat *stat)
+int ll_getattr(struct vfsmount *mnt, struct dentry *de, 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, MDS_INODELOCK_UPDATE |
-                                             MDS_INODELOCK_LOOKUP);
+       res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
+                                     MDS_INODELOCK_LOOKUP);
         ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
 
         if (res)
@@ -3287,15 +3355,9 @@ int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
 
         return 0;
 }
-int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
-{
-        struct lookup_intent it = { .it_op = IT_GETATTR };
-
-        return ll_getattr_it(mnt, de, &it, stat);
-}
 
-int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
-                __u64 start, __u64 len)
+static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
+                    __u64 start, __u64 len)
 {
         int rc;
         size_t num_bytes;
@@ -3313,21 +3375,24 @@ int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
         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));
+       if (extent_count > 0)
+               memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
+                      sizeof(struct ll_fiemap_extent));
 
-        rc = ll_do_fiemap(inode, fiemap, num_bytes);
+       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));
+       fieinfo->fi_flags = fiemap->fm_flags;
+       fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
+       if (extent_count > 0)
+               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;
+       OBD_FREE_LARGE(fiemap, num_bytes);
+       return rc;
 }
 
-struct posix_acl * ll_get_acl(struct inode *inode, int type)
+struct posix_acl *ll_get_acl(struct inode *inode, int type)
 {
        struct ll_inode_info *lli = ll_i2info(inode);
        struct posix_acl *acl = NULL;
@@ -3383,8 +3448,14 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
 # endif
 #endif
 {
-        int rc = 0;
-        ENTRY;
+       int rc = 0;
+       struct ll_sb_info *sbi;
+       struct root_squash_info *squash;
+       struct cred *cred = NULL;
+       const struct cred *old_cred = NULL;
+       cfs_cap_t cap;
+       bool squash_id = false;
+       ENTRY;
 
 #ifdef MAY_NOT_BLOCK
        if (mask & MAY_NOT_BLOCK)
@@ -3398,10 +3469,8 @@ 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_LOOKUP };
-
-                rc = __ll_inode_revalidate_it(inode->i_sb->s_root, &it,
-                                              MDS_INODELOCK_LOOKUP);
+               rc = __ll_inode_revalidate(inode->i_sb->s_root,
+                                          MDS_INODELOCK_LOOKUP);
                 if (rc)
                         RETURN(rc);
         }
@@ -3409,11 +3478,46 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
        CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
               PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
 
-       if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
-               return lustre_check_remote_perm(inode, mask);
+       /* squash fsuid/fsgid if needed */
+       sbi = ll_i2sbi(inode);
+       squash = &sbi->ll_squash;
+       if (unlikely(squash->rsi_uid != 0 &&
+                    uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
+                    !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
+                       squash_id = true;
+       }
+       if (squash_id) {
+               CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
+                      __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
+                      squash->rsi_uid, squash->rsi_gid);
+
+               /* update current process's credentials
+                * and FS capability */
+               cred = prepare_creds();
+               if (cred == NULL)
+                       RETURN(-ENOMEM);
+
+               cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
+               cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
+               for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
+                       if ((1 << cap) & CFS_CAP_FS_MASK)
+                               cap_lower(cred->cap_effective, cap);
+               }
+               old_cred = override_creds(cred);
+       }
+
+       ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
 
-       ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
-       rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
+       if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
+               rc = lustre_check_remote_perm(inode, mask);
+       else
+               rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
+
+       /* restore current process's credentials and FS capability */
+       if (squash_id) {
+               revert_creds(old_cred);
+               put_cred(cred);
+       }
 
        RETURN(rc);
 }
@@ -3486,15 +3590,15 @@ struct inode_operations ll_file_inode_operations = {
 /* dynamic ioctl number support routins */
 static struct llioc_ctl_data {
        struct rw_semaphore     ioc_sem;
-        cfs_list_t              ioc_head;
+       struct list_head        ioc_head;
 } llioc = {
-        __RWSEM_INITIALIZER(llioc.ioc_sem),
-        CFS_LIST_HEAD_INIT(llioc.ioc_head)
+       __RWSEM_INITIALIZER(llioc.ioc_sem),
+       LIST_HEAD_INIT(llioc.ioc_head)
 };
 
 
 struct llioc_data {
-        cfs_list_t              iocd_list;
+       struct list_head        iocd_list;
         unsigned int            iocd_size;
         llioc_callback_t        iocd_cb;
         unsigned int            iocd_count;
@@ -3523,7 +3627,7 @@ void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
         memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
 
        down_write(&llioc.ioc_sem);
-        cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
+       list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
        up_write(&llioc.ioc_sem);
 
         RETURN(in_data);
@@ -3537,11 +3641,11 @@ void ll_iocontrol_unregister(void *magic)
                 return;
 
        down_write(&llioc.ioc_sem);
-        cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
+       list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
                 if (tmp == magic) {
                         unsigned int size = tmp->iocd_size;
 
-                        cfs_list_del(&tmp->iocd_list);
+                       list_del(&tmp->iocd_list);
                        up_write(&llioc.ioc_sem);
 
                         OBD_FREE(tmp, size);
@@ -3556,15 +3660,16 @@ void ll_iocontrol_unregister(void *magic)
 EXPORT_SYMBOL(ll_iocontrol_register);
 EXPORT_SYMBOL(ll_iocontrol_unregister);
 
-enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file,
-                        unsigned int cmd, unsigned long arg, int *rcp)
+static enum llioc_iter
+ll_iocontrol_call(struct inode *inode, struct file *file,
+                 unsigned int cmd, unsigned long arg, int *rcp)
 {
         enum llioc_iter ret = LLIOC_CONT;
         struct llioc_data *data;
         int rc = -EINVAL, i;
 
        down_read(&llioc.ioc_sem);
-        cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
+       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;
@@ -3607,11 +3712,24 @@ int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
                LASSERT(lock != NULL);
                LASSERT(ldlm_has_layout(lock));
                if (result == 0) {
+                       struct lustre_md *md = conf->u.coc_md;
+                       __u32 gen = LL_LAYOUT_GEN_EMPTY;
+
                        /* it can only be allowed to match after layout is
                         * applied to inode otherwise false layout would be
                         * seen. Applying layout shoud happen before dropping
                         * the intent lock. */
                        ldlm_lock_allow_match(lock);
+
+                       lli->lli_has_smd = lsm_has_objects(md->lsm);
+                       if (md->lsm != NULL)
+                               gen = md->lsm->lsm_layout_gen;
+
+                       CDEBUG(D_VFSTRACE,
+                              DFID ": layout version change: %u -> %u\n",
+                              PFID(&lli->lli_fid), ll_layout_version_get(lli),
+                              gen);
+                       ll_layout_version_set(lli, gen);
                }
        }
        RETURN(result);
@@ -3644,7 +3762,7 @@ static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
         * layout here. Please note that we can't use the LVB buffer in
         * completion AST because it doesn't have a large enough buffer */
        oc = ll_mdscapa_get(inode);
-       rc = ll_get_max_mdsize(sbi, &lmmsize);
+       rc = ll_get_default_mdsize(sbi, &lmmsize);
        if (rc == 0)
                rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
                                OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
@@ -3654,10 +3772,10 @@ static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
                RETURN(rc);
 
        body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-       if (body == NULL || body->eadatasize > lmmsize)
+       if (body == NULL)
                GOTO(out, rc = -EPROTO);
 
-       lmmsize = body->eadatasize;
+       lmmsize = body->mbo_eadatasize;
        if (lmmsize == 0) /* empty layout */
                GOTO(out, rc = 0);
 
@@ -3708,7 +3826,7 @@ static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode,
        LASSERT(lock != NULL);
        LASSERT(ldlm_has_layout(lock));
 
-       LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured: %d\n",
+       LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured: %d",
                   PFID(&lli->lli_fid), inode, reconf);
 
        /* in case this is a caching lock and reinstate with new inode */
@@ -3725,7 +3843,7 @@ static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode,
                if (lvb_ready) {
                        /* layout_gen must be valid if layout lock is not
                         * cancelled and stripe has already set */
-                       *gen = lli->lli_layout_gen;
+                       *gen = ll_layout_version_get(lli);
                        rc = 0;
                }
                GOTO(out, rc);
@@ -3820,38 +3938,26 @@ int ll_layout_refresh(struct inode *inode, __u32 *gen)
        struct ldlm_enqueue_info einfo = {
                .ei_type = LDLM_IBITS,
                .ei_mode = LCK_CR,
-               .ei_cb_bl = ll_md_blocking_ast,
-               .ei_cb_cp = ldlm_completion_ast,
+               .ei_cb_bl = &ll_md_blocking_ast,
+               .ei_cb_cp = &ldlm_completion_ast,
        };
        int rc;
        ENTRY;
 
-       *gen = lli->lli_layout_gen;
-       if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
+       *gen = ll_layout_version_get(lli);
+       if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE)
                RETURN(0);
 
        /* sanity checks */
        LASSERT(fid_is_sane(ll_inode2fid(inode)));
        LASSERT(S_ISREG(inode->i_mode));
 
-       /* mostly layout lock is caching on the local side, so try to match
-        * it before grabbing layout lock mutex. */
-       mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
-                              LCK_CR | LCK_CW | LCK_PR | LCK_PW);
-       if (mode != 0) { /* hit cached lock */
-               rc = ll_layout_lock_set(&lockh, mode, inode, gen, false);
-               if (rc == 0)
-                       RETURN(0);
-
-               /* better hold lli_layout_mutex to try again otherwise
-                * it will have starvation problem. */
-       }
-
        /* take layout lock mutex to enqueue layout lock exclusively. */
        mutex_lock(&lli->lli_layout_mutex);
 
 again:
-       /* try again. Maybe somebody else has done this. */
+       /* mostly layout lock is caching on the local side, so try to match
+        * it before grabbing layout lock mutex. */
        mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
                               LCK_CR | LCK_CW | LCK_PR | LCK_PW);
        if (mode != 0) { /* hit cached lock */
@@ -3875,12 +3981,11 @@ again:
        it.it_op = IT_LAYOUT;
        lockh.cookie = 0ULL;
 
-       LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)\n",
+       LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
                          ll_get_fsname(inode->i_sb, NULL, 0),
                          PFID(&lli->lli_fid), inode);
 
-       rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh,
-                       NULL, 0, NULL, 0);
+       rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
        if (it.d.lustre.it_data != NULL)
                ptlrpc_req_finished(it.d.lustre.it_data);
        it.d.lustre.it_data = NULL;
@@ -3931,4 +4036,3 @@ int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
        OBD_FREE(hur, len);
        RETURN(rc);
 }
-