Whamcloud - gitweb
b=22187 Handle the NULL pointer as legal value.
[fs/lustre-release.git] / lustre / llite / dir.c
index b84b877..266ae99 100644 (file)
@@ -1,33 +1,44 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- * Copyright (C) 1992, 1993, 1994, 1995
- * Remy Card (card@masi.ibp.fr)
- * Laboratoire MASI - Institut Blaise Pascal
- * Universite Pierre et Marie Curie (Paris VI)
+ * GPL HEADER START
  *
- *  from
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *  linux/fs/minix/dir.c
- *  linux/fs/ext2/dir.c
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *  Copyright (C) 1991, 1992  Linus Torvalds
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *  ext2 directory handling functions
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
- *  Big-endian to little-endian byte-swapping/bitmaps by
- *        David S. Miller (davem@caip.rutgers.edu), 1995
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
  *
- *  All code that works with directory layout had been switched to pagecache
- *  and moved here. AV
+ * GPL HEADER END
+ */
+/*
+ * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  *
- *  Adapted for Lustre Light
- *  Copyright (C) 2002-2003, Cluster File Systems, Inc.
+ * lustre/llite/dir.c
  *
+ * Directory code for lustre client.
  */
 
 #include <linux/fs.h>
-#include <linux/ext2_fs.h>
 #include <linux/pagemap.h>
 #include <linux/mm.h>
 #include <linux/version.h>
 #include <lustre_fid.h>
 #include "llite_internal.h"
 
+#ifndef HAVE_PAGE_CHECKED
 #ifdef HAVE_PG_FS_MISC
 #define PageChecked(page)        test_bit(PG_fs_misc, &(page)->flags)
 #define SetPageChecked(page)     set_bit(PG_fs_misc, &(page)->flags)
+#else
+#error PageChecked or PageFsMisc not defined in kernel
+#endif
 #endif
 
 /*
  *
  */
 
-static __u32 hash_x_index(__u32 value)
-{
-        return ((__u32)~0) - value;
-}
-#ifdef HAVE_PG_FS_MISC
-#define PageChecked(page)        test_bit(PG_fs_misc, &(page)->flags)
-#define SetPageChecked(page)     set_bit(PG_fs_misc, &(page)->flags)
-#endif
-
 /* returns the page unlocked, but with a reference */
 static int ll_dir_readpage(struct file *file, struct page *page)
 {
@@ -149,7 +155,7 @@ static int ll_dir_readpage(struct file *file, struct page *page)
         int rc;
         ENTRY;
 
-        hash = hash_x_index(page->index);
+        hash = (__u64)hash_x_index(page->index);
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) off %lu\n",
                inode->i_ino, inode->i_generation, inode, (unsigned long)hash);
 
@@ -158,16 +164,12 @@ static int ll_dir_readpage(struct file *file, struct page *page)
                          oc, hash, page, &request);
         capa_put(oc);
         if (!rc) {
-                body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF,
-                                      sizeof(*body));
+                body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
                 /* Checked by mdc_readpage() */
                 LASSERT(body != NULL);
 
-                /* Swabbed by mdc_readpage() */
-                LASSERT_REPSWABBED(request, REPLY_REC_OFF);
-
                 if (body->valid & OBD_MD_FLSIZE)
-                        i_size_write(inode, body->size);
+                        cl_isize_write(inode, body->size);
                 SetPageUptodate(page);
         }
         ptlrpc_req_finished(request);
@@ -181,33 +183,34 @@ struct address_space_operations ll_dir_aops = {
         .readpage  = ll_dir_readpage,
 };
 
-static inline unsigned long dir_pages(struct inode *inode)
-{
-        return (i_size_read(inode) + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
-}
-
-static inline unsigned ll_chunk_size(struct inode *inode)
-{
-        return inode->i_sb->s_blocksize;
-}
-
 static void ll_check_page(struct inode *dir, struct page *page)
 {
         /* XXX: check page format later */
         SetPageChecked(page);
 }
 
-static inline void ll_put_page(struct page *page)
+static void ll_release_page(struct page *page, __u64 hash,
+                            __u64 start, __u64 end)
 {
         kunmap(page);
+        lock_page(page);
+        if (likely(page->mapping != NULL)) {
+                ll_truncate_complete_page(page);
+                unlock_page(page);
+        } else {
+                unlock_page(page);
+                CWARN("NULL mapping page %p, truncated by others: "
+                      "hash("LPX64") | start("LPX64") | end("LPX64")\n",
+                      page, hash, start, end);
+        }
         page_cache_release(page);
 }
 
 /*
  * Find, kmap and return page that contains given hash.
  */
-static struct page *ll_dir_page_locate(struct inode *dir, unsigned long hash,
-                                       __u32 *start, __u32 *end)
+static struct page *ll_dir_page_locate(struct inode *dir, __u64 hash,
+                                       __u64 *start, __u64 *end)
 {
         struct address_space *mapping = dir->i_mapping;
         /*
@@ -215,18 +218,18 @@ static struct page *ll_dir_page_locate(struct inode *dir, unsigned long hash,
          * radix_tree_gang_lookup() can be used to find a page with starting
          * hash _smaller_ than one we are looking for.
          */
-        unsigned long offset = hash_x_index(hash);
+        unsigned long offset = hash_x_index((unsigned long)hash);
         struct page *page;
         int found;
 
-       spin_lock_irq(&mapping->tree_lock);
-       found = radix_tree_gang_lookup(&mapping->page_tree,
+        TREE_READ_LOCK_IRQ(mapping);
+        found = radix_tree_gang_lookup(&mapping->page_tree,
                                        (void **)&page, offset, 1);
-       if (found > 0) {
+        if (found > 0) {
                 struct lu_dirpage *dp;
 
-               page_cache_get(page);
-                spin_unlock_irq(&mapping->tree_lock);
+                page_cache_get(page);
+                TREE_READ_UNLOCK_IRQ(mapping);
                 /*
                  * In contrast to find_lock_page() we are sure that directory
                  * page cannot be truncated (while DLM lock is held) and,
@@ -238,15 +241,11 @@ static struct page *ll_dir_page_locate(struct inode *dir, unsigned long hash,
                 wait_on_page(page);
                 if (PageUptodate(page)) {
                         dp = kmap(page);
-                        *start = le32_to_cpu(dp->ldp_hash_start);
-                        *end   = le32_to_cpu(dp->ldp_hash_end);
+                        *start = le64_to_cpu(dp->ldp_hash_start);
+                        *end   = le64_to_cpu(dp->ldp_hash_end);
                         LASSERT(*start <= hash);
                         if (hash > *end || (*end != *start && hash == *end)) {
-                                kunmap(page);
-                                lock_page(page);
-                                ll_truncate_complete_page(page);
-                                unlock_page(page);
-                                page_cache_release(page);
+                                ll_release_page(page, hash, *start, *end);
                                 page = NULL;
                         }
                 } else {
@@ -254,30 +253,15 @@ static struct page *ll_dir_page_locate(struct inode *dir, unsigned long hash,
                         page = ERR_PTR(-EIO);
                 }
 
-       } else {
-                spin_unlock_irq(&mapping->tree_lock);
+        } else {
+                TREE_READ_UNLOCK_IRQ(mapping);
                 page = NULL;
         }
         return page;
 }
 
-/*
- * Chain of hash overflow pages.
- */
-struct ll_dir_chain {
-        /* XXX something. Later */
-};
-
-static void ll_dir_chain_init(struct ll_dir_chain *chain)
-{
-}
-
-static void ll_dir_chain_fini(struct ll_dir_chain *chain)
-{
-}
-
-static struct page *ll_get_dir_page(struct inode *dir, __u32 hash, int exact,
-                                    struct ll_dir_chain *chain)
+struct page *ll_get_dir_page(struct inode *dir, __u64 hash, int exact,
+                             struct ll_dir_chain *chain)
 {
         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
         struct address_space *mapping = dir->i_mapping;
@@ -286,26 +270,27 @@ static struct page *ll_get_dir_page(struct inode *dir, __u32 hash, int exact,
         struct page *page;
         ldlm_mode_t mode;
         int rc;
-        __u32 start;
-        __u32 end;
+        __u64 start = 0;
+        __u64 end = 0;
 
         mode = LCK_PR;
         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
         if (!rc) {
                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
-                       ll_md_blocking_ast, ldlm_completion_ast, NULL, dir };
+                       ll_md_blocking_ast, ldlm_completion_ast,
+                       NULL, NULL, dir };
                 struct lookup_intent it = { .it_op = IT_READDIR };
                 struct ptlrpc_request *request;
                 struct md_op_data *op_data;
 
-                op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0, 
+                op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
                                              LUSTRE_OPC_ANY, NULL);
                 if (IS_ERR(op_data))
                         return (void *)op_data;
 
                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
-                                op_data, &lockh, NULL, 0, 0);
+                                op_data, &lockh, NULL, 0, NULL, 0);
 
                 ll_finish_md_op_data(op_data);
 
@@ -313,19 +298,24 @@ static struct page *ll_get_dir_page(struct inode *dir, __u32 hash, int exact,
                 if (request)
                         ptlrpc_req_finished(request);
                 if (rc < 0) {
-                        CERROR("lock enqueue: rc: %d\n", rc);
+                        CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
+                               PFID(ll_inode2fid(dir)), hash, rc);
                         return ERR_PTR(rc);
                 }
         } else {
                 /* for cross-ref object, l_ast_data of the lock may not be set,
                  * we reset it here */
-                md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie, dir);
+                md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
+                                 dir, NULL);
         }
         ldlm_lock_dump_handle(D_OTHER, &lockh);
 
         page = ll_dir_page_locate(dir, hash, &start, &end);
-        if (IS_ERR(page))
+        if (IS_ERR(page)) {
+                CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
+                       PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
                 GOTO(out_unlock, page);
+        }
 
         if (page != NULL) {
                 /*
@@ -349,34 +339,41 @@ static struct page *ll_get_dir_page(struct inode *dir, __u32 hash, int exact,
                          * entries with smaller hash values. Stale page should
                          * be invalidated, and new one fetched.
                          */
-                        CWARN("Stale readpage page %p: %#lx != %#lx\n", page,
-                              (unsigned long)hash, (unsigned long)start);
-                        lock_page(page);
-                        ll_truncate_complete_page(page);
-                        unlock_page(page);
-                        page_cache_release(page);
-                } else
+                        CDEBUG(D_OTHER, "Stale readpage page %p: "LPX64" != "LPX64"\n",
+                               page, hash, start);
+                        ll_release_page(page, hash, start, end);
+                } else {
                         GOTO(hash_collision, page);
+                }
         }
 
-        page = read_cache_page(mapping, hash_x_index(hash),
+        page = read_cache_page(mapping, hash_x_index((unsigned long)hash),
                                (filler_t*)mapping->a_ops->readpage, NULL);
-        if (IS_ERR(page))
+        if (IS_ERR(page)) {
+                CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
+                       PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
                 GOTO(out_unlock, page);
+        }
 
         wait_on_page(page);
         (void)kmap(page);
-        if (!PageUptodate(page))
+        if (!PageUptodate(page)) {
+                CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
+                       PFID(ll_inode2fid(dir)), hash, -5);
                 goto fail;
+        }
         if (!PageChecked(page))
                 ll_check_page(dir, page);
-        if (PageError(page))
+        if (PageError(page)) {
+                CERROR("page error: "DFID" at "LPU64": rc %d\n",
+                       PFID(ll_inode2fid(dir)), hash, -5);
                 goto fail;
+        }
 hash_collision:
         dp = page_address(page);
 
-        start = le32_to_cpu(dp->ldp_hash_start);
-        end   = le32_to_cpu(dp->ldp_hash_end);
+        start = le64_to_cpu(dp->ldp_hash_start);
+        end   = le64_to_cpu(dp->ldp_hash_end);
         if (end == start) {
                 LASSERT(start == hash);
                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
@@ -401,13 +398,13 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 {
         struct inode         *inode = filp->f_dentry->d_inode;
         struct ll_inode_info *info  = ll_i2info(inode);
-        struct ll_sb_info    *sbi   = ll_i2sbi(inode);
-        __u32                 pos   = filp->f_pos;
+        __u64                 pos   = filp->f_pos;
         struct page          *page;
         struct ll_dir_chain   chain;
         int rc;
         int done;
         int shift;
+        __u16 type;
         ENTRY;
 
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu\n",
@@ -432,12 +429,12 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
                 struct lu_dirent  *ent;
 
                 if (!IS_ERR(page)) {
-                        /* 
-                         * If page is empty (end of directoryis reached),
-                         * use this value. 
+                        /*
+                         * If page is empty (end of directory is reached),
+                         * use this value.
                          */
-                        __u32 hash = DIR_END_OFF;
-                        __u32 next;
+                        __u64 hash = DIR_END_OFF;
+                        __u64 next;
 
                         dp = page_address(page);
                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
@@ -451,7 +448,7 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
                                  * XXX: implement correct swabbing here.
                                  */
 
-                                hash    = le32_to_cpu(ent->lde_hash);
+                                hash    = le64_to_cpu(ent->lde_hash);
                                 namelen = le16_to_cpu(ent->lde_namelen);
 
                                 if (hash < pos)
@@ -470,12 +467,12 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
                                 fid  = ent->lde_fid;
                                 name = ent->lde_name;
                                 fid_le_to_cpu(&fid, &fid);
-                                ino  = ll_fid_build_ino(sbi, &fid);
-
+                                ino  = cl_fid_build_ino(&fid);
+                                type = ll_dirent_type_get(ent);
                                 done = filldir(cookie, name, namelen,
-                                               (loff_t)hash, ino, DT_UNKNOWN);
+                                               (loff_t)hash, ino, type);
                         }
-                        next = le32_to_cpu(dp->ldp_hash_end);
+                        next = le64_to_cpu(dp->ldp_hash_end);
                         ll_put_page(page);
                         if (!done) {
                                 pos = next;
@@ -505,7 +502,7 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
                 }
         }
 
-        filp->f_pos = (loff_t)(__s32)pos;
+        filp->f_pos = (loff_t)pos;
         filp->f_version = inode->i_version;
         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 
@@ -514,16 +511,6 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
         RETURN(rc);
 }
 
-#define QCTL_COPY(out, in)              \
-do {                                    \
-        Q_COPY(out, in, qc_cmd);        \
-        Q_COPY(out, in, qc_type);       \
-        Q_COPY(out, in, qc_id);         \
-        Q_COPY(out, in, qc_stat);       \
-        Q_COPY(out, in, qc_dqinfo);     \
-        Q_COPY(out, in, qc_dqblk);      \
-} while (0)
-
 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 {
         struct mgs_send_param *msp;
@@ -534,7 +521,7 @@ int ll_send_mgc_param(struct obd_export *mgc, char *string)
                 return -ENOMEM;
 
         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
-        rc = obd_set_info_async(mgc, strlen(KEY_SET_INFO), KEY_SET_INFO,
+        rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
                                 sizeof(struct mgs_send_param), msp, NULL);
         if (rc)
                 CERROR("Failed to set parameter: %d\n", rc);
@@ -570,17 +557,39 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
         struct obd_device *mgc = lsi->lsi_mgc;
         char *fsname = NULL, *param = NULL;
+        int lum_size;
 
-        /*
-         * This is coming from userspace, so should be in
-         * local endian.  But the MDS would like it in little
-         * endian, so we swab it before we send it.
-         */
-        if (lump->lmm_magic != LOV_USER_MAGIC)
-                RETURN(-EINVAL);
-
-        if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC))
-                lustre_swab_lov_user_md(lump);
+        if (lump != NULL) {
+                /*
+                 * This is coming from userspace, so should be in
+                 * local endian.  But the MDS would like it in little
+                 * endian, so we swab it before we send it.
+                 */
+                switch (lump->lmm_magic) {
+                case LOV_USER_MAGIC_V1: {
+                        if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
+                                lustre_swab_lov_user_md_v1(lump);
+                        lum_size = sizeof(struct lov_user_md_v1);
+                        break;
+                        }
+                case LOV_USER_MAGIC_V3: {
+                        if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
+                                lustre_swab_lov_user_md_v3(
+                                        (struct lov_user_md_v3 *)lump);
+                        lum_size = sizeof(struct lov_user_md_v3);
+                        break;
+                        }
+               default: {
+                        CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
+                                        " %#08x != %#08x nor %#08x\n",
+                                        lump->lmm_magic, LOV_USER_MAGIC_V1,
+                                        LOV_USER_MAGIC_V3);
+                        RETURN(-EINVAL);
+                        }
+               }
+        } else {
+                lum_size = sizeof(struct lov_user_md_v1);
+        }
 
         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
                                      LUSTRE_OPC_ANY, NULL);
@@ -588,7 +597,7 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
                 RETURN(PTR_ERR(op_data));
 
         /* swabbing is done in lov_setstripe() on server side */
-        rc = md_setattr(sbi->ll_md_exp, op_data, lump, sizeof(*lump),
+        rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
                         NULL, 0, &req, NULL);
         ll_finish_md_op_data(op_data);
         ptlrpc_req_finished(req);
@@ -597,6 +606,9 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
                         CERROR("mdc_setattr fails: rc = %d\n", rc);
         }
 
+        /* In the following we use the fact that LOV_USER_MAGIC_V1 and
+         LOV_USER_MAGIC_V3 have the same initial fields so we do not
+         need the make the distiction between the 2 versions */
         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 
@@ -604,21 +616,22 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
                 fsname = ll_get_fsname(inode);
                 /* Set root stripesize */
                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
-                        lump->lmm_stripe_size);
+                        lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
                 if (rc)
                         goto end;
 
                 /* Set root stripecount */
-                sprintf(param, "%s-MDT0000.lov.stripecount=%u", fsname,
-                        lump->lmm_stripe_count);
+                sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
+                        lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
                 if (rc)
                         goto end;
 
                 /* Set root stripeoffset */
-                sprintf(param, "%s-MDT0000.lov.stripeoffset=%u", fsname,
-                        lump->lmm_stripe_offset);
+                sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
+                        lump ? le16_to_cpu(lump->lmm_stripe_offset) :
+                        (typeof(lump->lmm_stripe_offset))(-1));
                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
                 if (rc)
                         goto end;
@@ -631,25 +644,29 @@ end:
         return rc;
 }
 
-int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp, 
-                     int *lmm_size, struct ptlrpc_request **request) 
+int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
+                     int *lmm_size, struct ptlrpc_request **request)
 {
         struct ll_sb_info *sbi = ll_i2sbi(inode);
         struct mdt_body   *body;
         struct lov_mds_md *lmm = NULL;
         struct ptlrpc_request *req = NULL;
         int rc, lmmsize;
-        struct obd_capa *oc;
-        
+        struct md_op_data *op_data;
+
         rc = ll_get_max_mdsize(sbi, &lmmsize);
         if (rc)
                 RETURN(rc);
 
-        oc = ll_mdscapa_get(inode);
-        rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode),
-                        oc, OBD_MD_FLEASIZE | OBD_MD_FLDIREA,
-                        lmmsize, &req);
-        capa_put(oc);
+        op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
+                                     0, lmmsize, LUSTRE_OPC_ANY,
+                                     NULL);
+        if (op_data == NULL)
+                RETURN(-ENOMEM);
+
+        op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
+        rc = md_getattr(sbi->ll_md_exp, op_data, &req);
+        ll_finish_md_op_data(op_data);
         if (rc < 0) {
                 CDEBUG(D_INFO, "md_getattr failed on inode "
                        "%lu/%u: rc %d\n", inode->i_ino,
@@ -657,10 +674,8 @@ int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
                 GOTO(out, rc);
         }
 
-        body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
-        LASSERT(body != NULL); /* checked by md_getattr_name */
-        /* swabbed by mdc_getattr_name */
-        LASSERT_REPSWABBED(req, REPLY_REC_OFF);
+        body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+        LASSERT(body != NULL);
 
         lmmsize = body->eadatasize;
 
@@ -669,18 +684,28 @@ int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
                 GOTO(out, rc = -ENODATA);
         }
 
-        lmm = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1, lmmsize);
+        lmm = req_capsule_server_sized_get(&req->rq_pill,
+                                           &RMF_MDT_MD, lmmsize);
         LASSERT(lmm != NULL);
-        LASSERT_REPSWABBED(req, REPLY_REC_OFF + 1);
 
         /*
          * This is coming from the MDS, so is probably in
          * little endian.  We convert it to host endian before
          * passing it to userspace.
          */
-        if (lmm->lmm_magic == __swab32(LOV_MAGIC)) {
-                lustre_swab_lov_user_md((struct lov_user_md *)lmm);
-                lustre_swab_lov_user_md_objects((struct lov_user_md *)lmm);
+        /* We don't swab objects for directories */
+        switch (le32_to_cpu(lmm->lmm_magic)) {
+        case LOV_MAGIC_V1:
+                if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
+                        lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
+                break;
+        case LOV_MAGIC_V3:
+                if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
+                        lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
+                break;
+        default:
+                CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
+                rc = -EPROTO;
         }
 out:
         *lmmp = lmm;
@@ -689,11 +714,55 @@ out:
         return rc;
 }
 
+/*
+ *  Get MDT index for the inode.
+ */
+int ll_get_mdt_idx(struct inode *inode)
+{
+        struct ll_sb_info *sbi = ll_i2sbi(inode);
+        struct md_op_data *op_data;
+        int rc, mdtidx;
+        ENTRY;
+
+        op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
+                                     0, LUSTRE_OPC_ANY, NULL);
+        if (op_data == NULL)
+                RETURN(-ENOMEM);
+
+        op_data->op_valid |= OBD_MD_MDTIDX;
+        rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
+        mdtidx = op_data->op_mds;
+        ll_finish_md_op_data(op_data);
+        if (rc < 0) {
+                CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
+                RETURN(rc);
+        }
+        return mdtidx;
+}
+
+static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
+{
+        void *ptr;
+        int rc;
+
+        OBD_ALLOC(ptr, len);
+        if (ptr == NULL)
+                return -ENOMEM;
+        if (cfs_copy_from_user(ptr, data, len)) {
+                OBD_FREE(ptr, len);
+                return -EFAULT;
+        }
+        rc = obd_iocontrol(cmd, exp, len, data, NULL);
+        OBD_FREE(ptr, len);
+        return rc;
+}
+
 static int ll_dir_ioctl(struct inode *inode, struct file *file,
                         unsigned int cmd, unsigned long arg)
 {
         struct ll_sb_info *sbi = ll_i2sbi(inode);
         struct obd_ioctl_data *data;
+        int rc = 0;
         ENTRY;
 
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
@@ -705,24 +774,36 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
 
         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
         switch(cmd) {
-        case EXT3_IOC_GETFLAGS:
-        case EXT3_IOC_SETFLAGS:
+        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));
         /* 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_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;
+        }
         case IOC_MDC_LOOKUP: {
                 struct ptlrpc_request *request = NULL;
-                int namelen, rc, len = 0;
+                int namelen, len = 0;
                 char *buf = NULL;
                 char *filename;
-                struct obd_capa *oc;
+                struct md_op_data *op_data;
 
                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
                 if (rc)
@@ -730,46 +811,56 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                 data = (void *)buf;
 
                 filename = data->ioc_inlbuf1;
-                namelen = data->ioc_inllen1;
+                namelen = strlen(filename);
 
                 if (namelen < 1) {
                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
-                        GOTO(out, rc = -EINVAL);
+                        GOTO(out_free, rc = -EINVAL);
                 }
 
-                oc = ll_mdscapa_get(inode);
-                rc = md_getattr_name(sbi->ll_md_exp, ll_inode2fid(inode), oc,
-                                     filename, namelen, OBD_MD_FLID, 0,
-                                     &request);
-                capa_put(oc);
+                op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
+                                             0, LUSTRE_OPC_ANY, NULL);
+                if (op_data == NULL)
+                        GOTO(out_free, rc = -ENOMEM);
+
+                op_data->op_valid = OBD_MD_FLID;
+                rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
+                ll_finish_md_op_data(op_data);
                 if (rc < 0) {
                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
-                        GOTO(out, rc);
+                        GOTO(out_free, rc);
                 }
-
                 ptlrpc_req_finished(request);
-
                 EXIT;
-        out:
+out_free:
                 obd_ioctl_freedata(buf, len);
                 return rc;
         }
         case LL_IOC_LOV_SETSTRIPE: {
-                struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
-                int rc = 0;
+                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 set_default = 0;
 
-                LASSERT(sizeof(lum) == sizeof(*lump));
-                LASSERT(sizeof(lum.lmm_objects[0]) ==
-                        sizeof(lump->lmm_objects[0]));
-                rc = copy_from_user(&lum, lump, sizeof(lum));
-                if (rc)
+                LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
+                LASSERT(sizeof(lumv3.lmm_objects[0]) ==
+                        sizeof(lumv3p->lmm_objects[0]));
+                /* first try with v1 which is smaller than v3 */
+                if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
                         RETURN(-EFAULT);
 
+                if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
+                        if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
+                                RETURN(-EFAULT);
+                }
+
                 if (inode->i_sb->s_root == file->f_dentry)
                         set_default = 1;
 
-                rc = ll_dir_setstripe(inode, &lum, set_default);
+                /* in v1 and v3 cases lumv1 points to data */
+                rc = ll_dir_setstripe(inode, lumv1, set_default);
 
                 RETURN(rc);
         }
@@ -784,7 +875,7 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                 struct lov_mds_md *lmm = NULL;
                 struct mdt_body *body;
                 char *filename = NULL;
-                int rc, lmmsize;
+                int lmmsize;
 
                 if (cmd == IOC_MDC_GETFILEINFO ||
                     cmd == IOC_MDC_GETFILESTRIPE) {
@@ -792,24 +883,22 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                         if (IS_ERR(filename))
                                 RETURN(PTR_ERR(filename));
 
-                        rc = ll_lov_getstripe_ea_info(inode, filename, &lmm, 
+                        rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
                                                       &lmmsize, &request);
                 } else {
                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
                 }
 
                 if (request) {
-                        body = lustre_msg_buf(request->rq_repmsg,
-                                              REPLY_REC_OFF, sizeof(*body));
-                        LASSERT(body != NULL); /* checked by md_getattr_name */
-                        /* swabbed by md_getattr_name */
-                        LASSERT_REPSWABBED(request, REPLY_REC_OFF);
+                        body = req_capsule_server_get(&request->rq_pill,
+                                                      &RMF_MDT_BODY);
+                        LASSERT(body != NULL);
                 } else {
                         GOTO(out_req, rc);
                 }
 
                 if (rc < 0) {
-                        if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO || 
+                        if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
                                                cmd == LL_IOC_MDC_GETINFO))
                                 GOTO(skip_lmm, rc = 0);
                         else
@@ -824,9 +913,8 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                         lmdp = (struct lov_user_mds_data *)arg;
                         lump = &lmdp->lmd_lmm;
                 }
-                rc = copy_to_user(lump, lmm, lmmsize);
-                if (rc)
-                        GOTO(out_lmm, rc = -EFAULT);
+                if (cfs_copy_to_user(lump, lmm, lmmsize))
+                        GOTO(out_req, rc = -EFAULT);
         skip_lmm:
                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
                         struct lov_user_mds_data *lmdp;
@@ -847,15 +935,11 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                         st.st_ino     = inode->i_ino;
 
                         lmdp = (struct lov_user_mds_data *)arg;
-                        rc = copy_to_user(&lmdp->lmd_st, &st, sizeof(st));
-                        if (rc)
-                                GOTO(out_lmm, rc = -EFAULT);
+                        if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
+                                GOTO(out_req, rc = -EFAULT);
                 }
 
                 EXIT;
-        out_lmm:
-                if (lmm && lmm->lmm_magic == LOV_MAGIC_JOIN)
-                        OBD_FREE(lmm, lmmsize);
         out_req:
                 ptlrpc_req_finished(request);
                 if (filename)
@@ -869,7 +953,6 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                 struct lov_mds_md *lmm;
                 int lmmsize;
                 lstat_t st;
-                int rc;
 
                 lumd = (struct lov_user_mds_data *)arg;
                 lum = &lumd->lmd_lmm;
@@ -879,18 +962,36 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                         RETURN(rc);
 
                 OBD_ALLOC(lmm, lmmsize);
-                rc = copy_from_user(lmm, lum, lmmsize);
-                if (rc)
+                if (cfs_copy_from_user(lmm, lum, lmmsize))
                         GOTO(free_lmm, rc = -EFAULT);
 
+                switch (lmm->lmm_magic) {
+                case LOV_USER_MAGIC_V1:
+                        if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
+                                break;
+                        /* swab objects first so that stripes num will be sane */
+                        lustre_swab_lov_user_md_objects(
+                                ((struct lov_user_md_v1 *)lmm)->lmm_objects,
+                                ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
+                        lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
+                        break;
+                case LOV_USER_MAGIC_V3:
+                        if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
+                                break;
+                        /* swab objects first so that stripes num will be sane */
+                        lustre_swab_lov_user_md_objects(
+                                ((struct lov_user_md_v3 *)lmm)->lmm_objects,
+                                ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
+                        lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
+                        break;
+                default:
+                        GOTO(free_lmm, rc = -EINVAL);
+                }
+
                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
                 if (rc < 0)
                         GOTO(free_lmm, rc = -ENOMEM);
 
-                rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp, lsm);
-                if (rc)
-                        GOTO(free_lsm, rc);
-
                 /* Perform glimpse_size operation. */
                 memset(&st, 0, sizeof(st));
 
@@ -898,8 +999,7 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                 if (rc)
                         GOTO(free_lsm, rc);
 
-                rc = copy_to_user(&lumd->lmd_st, &st, sizeof(st));
-                if (rc)
+                if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
                         GOTO(free_lsm, rc = -EFAULT);
 
                 EXIT;
@@ -911,11 +1011,9 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
         }
         case OBD_IOC_LLOG_CATINFO: {
                 struct ptlrpc_request *req = NULL;
-                char *buf = NULL;
-                int rc, len = 0;
-                char *bufs[3] = { NULL }, *str;
-                int lens[3] = { sizeof(struct ptlrpc_body) };
-                int size[2] = { sizeof(struct ptlrpc_body) };
+                char                  *buf = NULL;
+                char                  *str;
+                int                    len = 0;
 
                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
                 if (rc)
@@ -927,30 +1025,41 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                         RETURN(-EINVAL);
                 }
 
-                lens[REQ_REC_OFF] = data->ioc_inllen1;
-                bufs[REQ_REC_OFF] = data->ioc_inlbuf1;
-                if (data->ioc_inllen2) {
-                        lens[REQ_REC_OFF + 1] = data->ioc_inllen2;
-                        bufs[REQ_REC_OFF + 1] = data->ioc_inlbuf2;
-                } else {
-                        lens[REQ_REC_OFF + 1] = 0;
-                        bufs[REQ_REC_OFF + 1] = NULL;
+                req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
+                                           &RQF_LLOG_CATINFO);
+                if (req == NULL)
+                        GOTO(out_catinfo, rc = -ENOMEM);
+
+                req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
+                                     data->ioc_inllen1);
+                req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
+                                     data->ioc_inllen2);
+
+                rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
+                if (rc) {
+                        ptlrpc_request_free(req);
+                        GOTO(out_catinfo, rc);
                 }
 
-                req = ptlrpc_prep_req(sbi2mdc(sbi)->cl_import,
-                                      LUSTRE_LOG_VERSION, LLOG_CATINFO, 3, lens,
-                                      bufs);
-                if (!req)
-                        GOTO(out_catinfo, rc = -ENOMEM);
+                str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
+                memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
+                if (data->ioc_inllen2) {
+                        str = req_capsule_client_get(&req->rq_pill,
+                                                     &RMF_STRING);
+                        memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
+                }
 
-                size[REPLY_REC_OFF] = data->ioc_plen1;
-                ptlrpc_req_set_repsize(req, 2, size);
+                req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
+                                     data->ioc_plen1);
+                ptlrpc_request_set_replen(req);
 
                 rc = ptlrpc_queue_wait(req);
                 if (!rc) {
-                        str = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF,
-                                                data->ioc_plen1);
-                        rc = copy_to_user(data->ioc_pbuf1, str, data->ioc_plen1);
+                        str = req_capsule_server_get(&req->rq_pill,
+                                                     &RMF_STRING);
+                        if (cfs_copy_to_user(data->ioc_pbuf1, str,
+                                             data->ioc_plen1))
+                                rc = -EFAULT;
                 }
                 ptlrpc_req_finished(req);
         out_catinfo:
@@ -959,9 +1068,10 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
         }
         case OBD_IOC_QUOTACHECK: {
                 struct obd_quotactl *oqctl;
-                int rc, error = 0;
+                int error = 0;
 
-                if (!capable(CAP_SYS_ADMIN))
+                if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
+                    sbi->ll_flags & LL_SBI_RMT_CLIENT)
                         RETURN(-EPERM);
 
                 OBD_ALLOC_PTR(oqctl);
@@ -983,9 +1093,9 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
         }
         case OBD_IOC_POLL_QUOTACHECK: {
                 struct if_quotacheck *check;
-                int rc;
 
-                if (!capable(CAP_SYS_ADMIN))
+                if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
+                    sbi->ll_flags & LL_SBI_RMT_CLIENT)
                         RETURN(-EPERM);
 
                 OBD_ALLOC_PTR(check);
@@ -996,7 +1106,8 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                                    NULL);
                 if (rc) {
                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
-                        if (copy_to_user((void *)arg, check, sizeof(*check)))
+                        if (cfs_copy_to_user((void *)arg, check,
+                                             sizeof(*check)))
                                 rc = -EFAULT;
                         GOTO(out_poll, rc);
                 }
@@ -1005,7 +1116,8 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                                    NULL);
                 if (rc) {
                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
-                        if (copy_to_user((void *)arg, check, sizeof(*check)))
+                        if (cfs_copy_to_user((void *)arg, check,
+                                             sizeof(*check)))
                                 rc = -EFAULT;
                         GOTO(out_poll, rc);
                 }
@@ -1013,47 +1125,39 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                 OBD_FREE_PTR(check);
                 RETURN(rc);
         }
-#ifdef HAVE_QUOTA_SUPPORT
         case OBD_IOC_QUOTACTL: {
                 struct if_quotactl *qctl;
-                struct obd_quotactl *oqctl;
-
-                int cmd, type, id, rc = 0;
+                int cmd, type, id, valid;
 
                 OBD_ALLOC_PTR(qctl);
                 if (!qctl)
                         RETURN(-ENOMEM);
 
-                OBD_ALLOC_PTR(oqctl);
-                if (!oqctl) {
-                        OBD_FREE_PTR(qctl);
-                        RETURN(-ENOMEM);
-                }
-                if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
+                if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
                         GOTO(out_quotactl, rc = -EFAULT);
 
                 cmd = qctl->qc_cmd;
                 type = qctl->qc_type;
                 id = qctl->qc_id;
+                valid = qctl->qc_valid;
+
                 switch (cmd) {
+                case LUSTRE_Q_INVALIDATE:
+                case LUSTRE_Q_FINVALIDATE:
                 case Q_QUOTAON:
                 case Q_QUOTAOFF:
                 case Q_SETQUOTA:
                 case Q_SETINFO:
-                        if (!capable(CAP_SYS_ADMIN))
+                        if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
+                            sbi->ll_flags & LL_SBI_RMT_CLIENT)
                                 GOTO(out_quotactl, rc = -EPERM);
                         break;
                 case Q_GETQUOTA:
-                        if (((type == USRQUOTA && current->euid != id) ||
+                        if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
                              (type == GRPQUOTA && !in_egroup_p(id))) &&
-                            !capable(CAP_SYS_ADMIN))
+                            (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
+                             sbi->ll_flags & LL_SBI_RMT_CLIENT))
                                 GOTO(out_quotactl, rc = -EPERM);
-
-                        /* XXX: dqb_valid is borrowed as a flag to mark that
-                         *      only mds quota is wanted */
-                        if (qctl->qc_dqblk.dqb_valid)
-                                qctl->obd_uuid = sbi->ll_md_exp->exp_obd->
-                                                        u.cli.cl_target_uuid;
                         break;
                 case Q_GETINFO:
                         break;
@@ -1062,96 +1166,140 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
                         GOTO(out_quotactl, rc = -ENOTTY);
                 }
 
-                QCTL_COPY(oqctl, qctl);
-
-                if (qctl->obd_uuid.uuid[0]) {
-                        struct obd_device *obd;
-                        struct obd_uuid *uuid = &qctl->obd_uuid;
-
-                        obd = class_find_client_notype(uuid,
-                                         &sbi->ll_dt_exp->exp_obd->obd_uuid);
-                        if (!obd)
-                                GOTO(out_quotactl, rc = -ENOENT);
+                if (valid != QC_GENERAL) {
+                        if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
+                                GOTO(out_quotactl, rc = -EOPNOTSUPP);
 
                         if (cmd == Q_GETINFO)
-                                oqctl->qc_cmd = Q_GETOINFO;
+                                qctl->qc_cmd = Q_GETOINFO;
                         else if (cmd == Q_GETQUOTA)
-                                oqctl->qc_cmd = Q_GETOQUOTA;
+                                qctl->qc_cmd = Q_GETOQUOTA;
                         else
                                 GOTO(out_quotactl, rc = -EINVAL);
 
-                        if (sbi->ll_md_exp->exp_obd == obd) {
-                                rc = obd_quotactl(sbi->ll_md_exp, oqctl);
-                        } else {
-                                int i;
-                                struct obd_export *exp;
-                                struct lov_obd *lov = &sbi->ll_dt_exp->
-                                                            exp_obd->u.lov;
-
-                                for (i = 0; i < lov->desc.ld_tgt_count; i++) {
-                                        if (!lov->lov_tgts[i] ||
-                                            !lov->lov_tgts[i]->ltd_active)
-                                                continue;
-                                        exp = lov->lov_tgts[i]->ltd_exp;
-                                        if (exp->exp_obd == obd) {
-                                                rc = obd_quotactl(exp, oqctl);
-                                                break;
-                                        }
-                                }
+                        switch (valid) {
+                        case QC_MDTIDX:
+                                rc = obd_iocontrol(OBD_IOC_QUOTACTL,
+                                                   sbi->ll_md_exp,
+                                                   sizeof(*qctl), qctl, NULL);
+                                break;
+                        case QC_OSTIDX:
+                                rc = obd_iocontrol(OBD_IOC_QUOTACTL,
+                                                   sbi->ll_dt_exp,
+                                                   sizeof(*qctl), qctl, NULL);
+                                break;
+                        case QC_UUID:
+                                rc = obd_iocontrol(OBD_IOC_QUOTACTL,
+                                                   sbi->ll_md_exp,
+                                                   sizeof(*qctl), qctl, NULL);
+                                if (rc == -EAGAIN)
+                                        rc = obd_iocontrol(OBD_IOC_QUOTACTL,
+                                                           sbi->ll_dt_exp,
+                                                           sizeof(*qctl), qctl,
+                                                           NULL);
+                                break;
+                        default:
+                                rc = -EINVAL;
+                                break;
                         }
 
-                        oqctl->qc_cmd = cmd;
-                        QCTL_COPY(qctl, oqctl);
-
-                        if (copy_to_user((void *)arg, qctl, sizeof(*qctl)))
-                                rc = -EFAULT;
-
-                        GOTO(out_quotactl, rc);
-                }
-
-                rc = obd_quotactl(sbi->ll_md_exp, oqctl);
-                if (rc && rc != -EBUSY && cmd == Q_QUOTAON) {
-                        oqctl->qc_cmd = Q_QUOTAOFF;
-                        obd_quotactl(sbi->ll_md_exp, oqctl);
+                        if (rc)
+                                GOTO(out_quotactl, rc);
+                        else
+                                qctl->qc_cmd = cmd;
+                } else {
+                        struct obd_quotactl *oqctl;
+
+                        OBD_ALLOC_PTR(oqctl);
+                        if (!oqctl)
+                                GOTO(out_quotactl, rc = -ENOMEM);
+
+                        QCTL_COPY(oqctl, qctl);
+                        rc = obd_quotactl(sbi->ll_md_exp, oqctl);
+                        if (rc) {
+                                if (rc != -EALREADY && cmd == Q_QUOTAON) {
+                                        oqctl->qc_cmd = Q_QUOTAOFF;
+                                        obd_quotactl(sbi->ll_md_exp, oqctl);
+                                }
+                                OBD_FREE_PTR(oqctl);
+                                GOTO(out_quotactl, rc);
+                        } else {
+                                QCTL_COPY(qctl, oqctl);
+                                OBD_FREE_PTR(oqctl);
+                        }
                 }
 
-                QCTL_COPY(qctl, oqctl);
-
-                if (copy_to_user((void *)arg, qctl, sizeof(*qctl)))
+                if (cfs_copy_to_user((void *)arg, qctl, sizeof(*qctl)))
                         rc = -EFAULT;
+
         out_quotactl:
                 OBD_FREE_PTR(qctl);
-                OBD_FREE_PTR(oqctl);
                 RETURN(rc);
         }
-#endif /* HAVE_QUOTA_SUPPORT */
         case OBD_IOC_GETNAME: {
                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
                 if (!obd)
                         RETURN(-EFAULT);
-                if (copy_to_user((void *)arg, obd->obd_name,
-                                strlen(obd->obd_name) + 1))
+                if (cfs_copy_to_user((void *)arg, obd->obd_name,
+                                     strlen(obd->obd_name) + 1))
                         RETURN (-EFAULT);
                 RETURN(0);
         }
         case LL_IOC_FLUSHCTX:
                 RETURN(ll_flush_ctx(inode));
-        case LL_IOC_GETFACL: {
-                struct rmtacl_ioctl_data ioc;
+#ifdef CONFIG_FS_POSIX_ACL
+        case LL_IOC_RMTACL: {
+            if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
+                inode == inode->i_sb->s_root->d_inode) {
+                struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
+
+                LASSERT(fd != NULL);
+                rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
+                if (!rc)
+                        fd->fd_flags |= LL_FILE_RMTACL;
+                RETURN(rc);
+            } else
+                RETURN(0);
+        }
+#endif
+        case LL_IOC_GETOBDCOUNT: {
+                int count;
 
-                if (copy_from_user(&ioc, (void *)arg, sizeof(ioc)))
+                if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
                         RETURN(-EFAULT);
 
-                RETURN(ll_ioctl_getfacl(inode, &ioc));
-        }
-        case LL_IOC_SETFACL: {
-                struct rmtacl_ioctl_data ioc;
+                if (!count) {
+                        /* get ost count */
+                        struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
+                        count = lov->desc.ld_tgt_count;
+                } else {
+                        /* get mdt count */
+                        struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
+                        count = lmv->desc.ld_tgt_count;
+                }
 
-                if (copy_from_user(&ioc, (void *)arg, sizeof(ioc)))
+                if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
                         RETURN(-EFAULT);
 
-                RETURN(ll_ioctl_setfacl(inode, &ioc));
+                RETURN(0);
         }
+        case LL_IOC_PATH2FID:
+                if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
+                                     sizeof(struct lu_fid)))
+                        RETURN(-EFAULT);
+                RETURN(0);
+        case OBD_IOC_CHANGELOG_SEND:
+        case OBD_IOC_CHANGELOG_CLEAR:
+                rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
+                                    sizeof(struct ioc_changelog));
+                RETURN(rc);
+        case OBD_IOC_FID2PATH:
+                RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
+        case LL_IOC_HSM_CT_START:
+                rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
+                                    sizeof(struct lustre_kernelcomm));
+                RETURN(rc);
+
         default:
                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
         }
@@ -1176,4 +1324,3 @@ struct file_operations ll_dir_operations = {
         .readdir  = ll_readdir,
         .ioctl    = ll_dir_ioctl
 };
-