Whamcloud - gitweb
LU-4984 llite: check for integer overflow in hsm user request
[fs/lustre-release.git] / lustre / llite / dir.c
index e9c35ba..cea0dd3 100644 (file)
@@ -51,8 +51,8 @@
 #include <lustre/lustre_idl.h>
 #include <obd_support.h>
 #include <obd_class.h>
+#include <lustre_ioctl.h>
 #include <lustre_lib.h>
-#include <lustre/lustre_idl.h>
 #include <lustre_lite.h>
 #include <lustre_dlm.h>
 #include <lustre_fid.h>
  * lmv_adjust_dirpages().
  *
  */
-
-/* returns the page unlocked, but with a reference */
-static int ll_dir_filler(void *_hash, struct page *page0)
+struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
+                            __u64 offset, struct ll_dir_chain *chain)
 {
-        struct inode *inode = page0->mapping->host;
-        int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
-        struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
-        struct ptlrpc_request *request;
-        struct mdt_body *body;
-        struct md_op_data *op_data;
-       __u64 hash = *((__u64 *)_hash);
-        struct page **page_pool;
-        struct page *page;
-        struct lu_dirpage *dp;
-       int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
-        int nrdpgs = 0; /* number of pages read actually */
-        int npages;
-        int i;
-        int rc;
-        ENTRY;
-
-        CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
-               inode->i_ino, inode->i_generation, inode, hash);
-
-       LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
-
-        OBD_ALLOC(page_pool, sizeof(page) * max_pages);
-        if (page_pool != NULL) {
-                page_pool[0] = page0;
-        } else {
-                page_pool = &page0;
-                max_pages = 1;
-        }
-        for (npages = 1; npages < max_pages; npages++) {
-                page = page_cache_alloc_cold(inode->i_mapping);
-                if (!page)
-                        break;
-                page_pool[npages] = page;
-        }
-
-        op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
-                                     LUSTRE_OPC_ANY, NULL);
-        op_data->op_npages = npages;
-        op_data->op_offset = hash;
-        rc = md_readpage(exp, op_data, page_pool, &request);
-        ll_finish_md_op_data(op_data);
-        if (rc == 0) {
-                body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
-                /* Checked by mdc_readpage() */
-                LASSERT(body != NULL);
-
-                if (body->valid & OBD_MD_FLSIZE)
-                        cl_isize_write(inode, body->size);
-
-               nrdpgs = (request->rq_bulk->bd_nob_transferred +
-                         PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-                SetPageUptodate(page0);
-        }
-        unlock_page(page0);
-        ptlrpc_req_finished(request);
-
-        CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
-
-        for (i = 1; i < npages; i++) {
-                unsigned long offset;
-                int ret;
-
-                page = page_pool[i];
-
-                if (rc < 0 || i >= nrdpgs) {
-                        page_cache_release(page);
-                        continue;
-                }
-
-                SetPageUptodate(page);
-
-               dp = kmap(page);
-               hash = le64_to_cpu(dp->ldp_hash_start);
-               kunmap(page);
+       struct md_callback      cb_op;
+       struct page             *page;
+       int                     rc;
 
-                offset = hash_x_index(hash, hash64);
+       cb_op.md_blocking_ast = ll_md_blocking_ast;
+       rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
+       if (rc != 0)
+               return ERR_PTR(rc);
 
-               prefetchw(&page->flags);
-               ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
-                                           GFP_KERNEL);
-               if (ret == 0)
-                        unlock_page(page);
-               else
-                        CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
-                               " %d\n", offset, ret);
-                page_cache_release(page);
-        }
-
-        if (page_pool != &page0)
-                OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
-        EXIT;
-        return rc;
+       return page;
 }
 
-static void ll_check_page(struct inode *dir, struct page *page)
+void ll_release_page(struct inode *inode, struct page *page,
+                    bool remove)
 {
-        /* XXX: check page format later */
-        SetPageChecked(page);
-}
+       kunmap(page);
+
+       /* Always remove the page for striped dir, because the page is
+        * built from temporarily in LMV layer */
+       if (inode != NULL && S_ISDIR(inode->i_mode) &&
+           ll_i2info(inode)->lli_lsm_md != NULL) {
+               __free_page(page);
+               return;
+       }
 
-void ll_release_page(struct page *page, int remove)
-{
-        kunmap(page);
-        if (remove) {
-                lock_page(page);
-                if (likely(page->mapping != NULL))
-                        truncate_complete_page(page->mapping, page);
-                unlock_page(page);
-        }
-        page_cache_release(page);
+       if (remove) {
+               lock_page(page);
+               if (likely(page->mapping != NULL))
+                       truncate_complete_page(page->mapping, page);
+               unlock_page(page);
+       }
+       page_cache_release(page);
 }
 
-/*
- * Find, kmap and return page that contains given hash.
- */
-static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
-                                       __u64 *start, __u64 *end)
+#ifdef HAVE_DIR_CONTEXT
+int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
+               struct dir_context *ctx)
 {
-        int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
-        struct address_space *mapping = dir->i_mapping;
-        /*
-         * Complement of hash is used as an index so that
-         * 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, hash64);
-        struct page *page;
-        int found;
-
-        TREE_READ_LOCK_IRQ(mapping);
-        found = radix_tree_gang_lookup(&mapping->page_tree,
-                                       (void **)&page, offset, 1);
-        if (found > 0) {
-                struct lu_dirpage *dp;
-
-                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,
-                 * hence, can avoid restart.
-                 *
-                 * In fact, page cannot be locked here at all, because
-                * ll_dir_filler() does synchronous io.
-                 */
-               wait_on_page_locked(page);
-               if (PageUptodate(page)) {
-                       dp = kmap(page);
-                        if (BITS_PER_LONG == 32 && hash64) {
-                                *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
-                                *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
-                                *hash  = *hash >> 32;
-                        } else {
-                                *start = le64_to_cpu(dp->ldp_hash_start);
-                                *end   = le64_to_cpu(dp->ldp_hash_end);
-                        }
-                        LASSERTF(*start <= *hash, "start = "LPX64",end = "
-                                 LPX64",hash = "LPX64"\n", *start, *end, *hash);
-                        CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
-                               offset, *start, *end, *hash);
-                        if (*hash > *end) {
-                                ll_release_page(page, 0);
-                                page = NULL;
-                        } else if (*end != *start && *hash == *end) {
-                                /*
-                                 * upon hash collision, remove this page,
-                                 * otherwise put page reference, and
-                                 * ll_get_dir_page() will issue RPC to fetch
-                                 * the page we want.
-                                 */
-                                ll_release_page(page,
-                                    le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
-                                page = NULL;
-                        }
-                } else {
-                        page_cache_release(page);
-                        page = ERR_PTR(-EIO);
-                }
-
-        } else {
-                TREE_READ_UNLOCK_IRQ(mapping);
-                page = NULL;
-        }
-        return page;
-}
-
-struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
-                             struct ll_dir_chain *chain)
+#else
+int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
+               void *cookie, filldir_t filldir)
 {
-        ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
-        struct address_space *mapping = dir->i_mapping;
-        struct lustre_handle lockh;
-        struct lu_dirpage *dp;
-        struct page *page;
-        ldlm_mode_t mode;
-        int rc;
-        __u64 start = 0;
-        __u64 end = 0;
-        __u64 lhash = hash;
-        struct ll_inode_info *lli = ll_i2info(dir);
-        int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
-
-        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 = {
-                       .ei_type = LDLM_IBITS,
-                       .ei_mode = mode,
-                       .ei_cb_bl = ll_md_blocking_ast,
-                       .ei_cb_cp = ldlm_completion_ast,
-               };
-               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,
-               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, NULL, 0);
-
-               ll_finish_md_op_data(op_data);
-
-               request = (struct ptlrpc_request *)it.d.lustre.it_data;
-               if (request)
-                       ptlrpc_req_finished(request);
-               if (rc < 0) {
-                       CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
-                               PFID(ll_inode2fid(dir)), hash, rc);
-                       return ERR_PTR(rc);
-               }
+#endif
+       struct ll_sb_info    *sbi        = ll_i2sbi(inode);
+       __u64                 pos        = *ppos;
+       bool                  is_api32 = ll_need_32bit_api(sbi);
+       bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
+       struct page          *page;
+       struct ll_dir_chain   chain;
+       bool                  done = false;
+       int                   rc = 0;
+       ENTRY;
 
-               CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
-                      dir, dir->i_ino, dir->i_generation);
-               md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
-                                &it.d.lustre.it_lock_handle, dir, NULL);
-        } 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, NULL);
-        }
-        ldlm_lock_dump_handle(D_OTHER, &lockh);
-
-       mutex_lock(&lli->lli_readdir_mutex);
-        page = ll_dir_page_locate(dir, &lhash, &start, &end);
-        if (IS_ERR(page)) {
-                CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
-                       PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
-                GOTO(out_unlock, page);
-        } else if (page != NULL) {
-                /*
-                 * XXX nikita: not entirely correct handling of a corner case:
-                 * suppose hash chain of entries with hash value HASH crosses
-                 * border between pages P0 and P1. First both P0 and P1 are
-                 * cached, seekdir() is called for some entry from the P0 part
-                 * of the chain. Later P0 goes out of cache. telldir(HASH)
-                 * happens and finds P1, as it starts with matching hash
-                 * value. Remaining entries from P0 part of the chain are
-                 * skipped. (Is that really a bug?)
-                 *
-                 * Possible solutions: 0. don't cache P1 is such case, handle
-                 * it as an "overflow" page. 1. invalidate all pages at
-                 * once. 2. use HASH|1 as an index for P1.
-                 */
-                GOTO(hash_collision, page);
-        }
+       ll_dir_chain_init(&chain);
 
-        page = read_cache_page(mapping, hash_x_index(hash, hash64),
-                              ll_dir_filler, &lhash);
-        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);
-        }
+       page = ll_get_dir_page(inode, op_data, pos, &chain);
 
-       wait_on_page_locked(page);
-       (void)kmap(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)) {
-                CERROR("page error: "DFID" at "LPU64": rc %d\n",
-                       PFID(ll_inode2fid(dir)), hash, -5);
-                goto fail;
-        }
-hash_collision:
-        dp = page_address(page);
-        if (BITS_PER_LONG == 32 && hash64) {
-                start = le64_to_cpu(dp->ldp_hash_start) >> 32;
-                end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
-                lhash = hash >> 32;
-        } else {
-                start = le64_to_cpu(dp->ldp_hash_start);
-                end   = le64_to_cpu(dp->ldp_hash_end);
-                lhash = hash;
-        }
-        if (end == start) {
-                LASSERT(start == lhash);
-                CWARN("Page-wide hash collision: "LPU64"\n", end);
-                if (BITS_PER_LONG == 32 && hash64)
-                        CWARN("Real page-wide hash collision at ["LPU64" "LPU64
-                              "] with hash "LPU64"\n",
-                              le64_to_cpu(dp->ldp_hash_start),
-                              le64_to_cpu(dp->ldp_hash_end), hash);
-                /*
-                 * Fetch whole overflow chain...
-                 *
-                 * XXX not yet.
-                 */
-                goto fail;
-        }
-out_unlock:
-       mutex_unlock(&lli->lli_readdir_mutex);
-        ldlm_lock_decref(&lockh, mode);
-        return page;
-
-fail:
-        ll_release_page(page, 1);
-        page = ERR_PTR(-EIO);
-        goto out_unlock;
-}
+       while (rc == 0 && !done) {
+               struct lu_dirpage *dp;
+               struct lu_dirent  *ent;
+               __u64 hash;
+               __u64 next;
 
-int ll_dir_read(struct inode *inode, __u64 *_pos, void *cookie,
-               filldir_t filldir)
-{
-        struct ll_inode_info *info       = ll_i2info(inode);
-        struct ll_sb_info    *sbi        = ll_i2sbi(inode);
-       __u64                 pos        = *_pos;
-        int                   api32      = ll_need_32bit_api(sbi);
-        int                   hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
-        struct page          *page;
-        struct ll_dir_chain   chain;
-       int                   done = 0;
-       int                   rc = 0;
-        ENTRY;
+               if (IS_ERR(page)) {
+                       rc = PTR_ERR(page);
+                       break;
+               }
 
-        ll_dir_chain_init(&chain);
-
-       page = ll_get_dir_page(inode, pos, &chain);
-
-        while (rc == 0 && !done) {
-                struct lu_dirpage *dp;
-                struct lu_dirent  *ent;
-
-                if (!IS_ERR(page)) {
-                        /*
-                         * If page is empty (end of directory is reached),
-                         * use this value.
-                         */
-                        __u64 hash = MDS_DIR_END_OFF;
-                        __u64 next;
-
-                        dp = page_address(page);
-                        for (ent = lu_dirent_start(dp); ent != NULL && !done;
-                             ent = lu_dirent_next(ent)) {
-                                __u16          type;
-                                int            namelen;
-                                struct lu_fid  fid;
-                                __u64          lhash;
-                                __u64          ino;
-
-                                /*
-                                 * XXX: implement correct swabbing here.
-                                 */
-
-                                hash = le64_to_cpu(ent->lde_hash);
-                                if (hash < pos)
-                                        /*
-                                         * Skip until we find target hash
-                                         * value.
-                                         */
-                                        continue;
-
-                                namelen = le16_to_cpu(ent->lde_namelen);
-                                if (namelen == 0)
-                                        /*
-                                         * Skip dummy record.
-                                         */
-                                        continue;
-
-                                if (api32 && hash64)
-                                        lhash = hash >> 32;
-                                else
-                                        lhash = hash;
-                                fid_le_to_cpu(&fid, &ent->lde_fid);
-                                ino = cl_fid_build_ino(&fid, api32);
-                                type = ll_dirent_type_get(ent);
-                                /* For 'll_nfs_get_name_filldir()', it will try
-                                 * to access the 'ent' through its 'lde_name',
-                                 * so the parameter 'name' for 'filldir()' must
-                                 * be part of the 'ent'. */
-                                done = filldir(cookie, ent->lde_name, namelen,
-                                               lhash, ino, type);
-                        }
-                        next = le64_to_cpu(dp->ldp_hash_end);
-                        if (!done) {
-                                pos = next;
-                                if (pos == MDS_DIR_END_OFF) {
-                                        /*
-                                         * End of directory reached.
-                                         */
-                                        done = 1;
-                                        ll_release_page(page, 0);
-                                } else if (1 /* chain is exhausted*/) {
-                                        /*
-                                         * Normal case: continue to the next
-                                         * page.
-                                         */
-                                        ll_release_page(page,
-                                            le32_to_cpu(dp->ldp_flags) &
-                                                        LDF_COLLIDE);
-                                       next = pos;
-                                       page = ll_get_dir_page(inode, pos,
-                                                               &chain);
-                                } else {
-                                        /*
-                                         * go into overflow page.
-                                         */
-                                        LASSERT(le32_to_cpu(dp->ldp_flags) &
-                                                LDF_COLLIDE);
-                                        ll_release_page(page, 1);
-                                }
-                        } else {
-                                pos = hash;
-                                ll_release_page(page, 0);
-                        }
-                } else {
-                        rc = PTR_ERR(page);
-                        CERROR("error reading dir "DFID" at %lu: rc %d\n",
-                               PFID(&info->lli_fid), (unsigned long)pos, rc);
-                }
-        }
+               hash = MDS_DIR_END_OFF;
+               dp = page_address(page);
+               for (ent = lu_dirent_start(dp); ent != NULL && !done;
+                    ent = lu_dirent_next(ent)) {
+                       __u16          type;
+                       int            namelen;
+                       struct lu_fid  fid;
+                       __u64          lhash;
+                       __u64          ino;
+
+                       hash = le64_to_cpu(ent->lde_hash);
+                       if (hash < pos)
+                               /*
+                                * Skip until we find target hash
+                                * value.
+                                */
+                               continue;
+
+                       namelen = le16_to_cpu(ent->lde_namelen);
+                       if (namelen == 0)
+                               /*
+                                * Skip dummy record.
+                                */
+                               continue;
+
+                       if (is_api32 && is_hash64)
+                               lhash = hash >> 32;
+                       else
+                               lhash = hash;
+                       fid_le_to_cpu(&fid, &ent->lde_fid);
+                       ino = cl_fid_build_ino(&fid, is_api32);
+                       type = ll_dirent_type_get(ent);
+                       /* For 'll_nfs_get_name_filldir()', it will try
+                        * to access the 'ent' through its 'lde_name',
+                        * so the parameter 'name' for 'filldir()' must
+                        * be part of the 'ent'. */
+#ifdef HAVE_DIR_CONTEXT
+                       ctx->pos = lhash;
+                       done = !dir_emit(ctx, ent->lde_name, namelen, ino,
+                                        type);
+#else
+                       done = filldir(cookie, ent->lde_name, namelen, lhash,
+                                      ino, type);
+#endif
+               }
+
+               if (done) {
+                       pos = hash;
+                       ll_release_page(inode, page, false);
+                       break;
+               }
 
-       *_pos = pos;
+               next = le64_to_cpu(dp->ldp_hash_end);
+               pos = next;
+               if (pos == MDS_DIR_END_OFF) {
+                       /*
+                        * End of directory reached.
+                        */
+                       done = 1;
+                       ll_release_page(inode, page, false);
+               } else {
+                       /*
+                        * Normal case: continue to the next
+                        * page.
+                        */
+                       ll_release_page(inode, page,
+                                       le32_to_cpu(dp->ldp_flags) &
+                                       LDF_COLLIDE);
+                       next = pos;
+                       page = ll_get_dir_page(inode, op_data, pos,
+                                              &chain);
+               }
+       }
+#ifdef HAVE_DIR_CONTEXT
+       ctx->pos = pos;
+#else
+       *ppos = pos;
+#endif
        ll_dir_chain_fini(&chain);
        RETURN(rc);
 }
 
+#ifdef HAVE_DIR_CONTEXT
+static int ll_iterate(struct file *filp, struct dir_context *ctx)
+#else
 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
+#endif
 {
        struct inode            *inode  = filp->f_dentry->d_inode;
        struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
        struct ll_sb_info       *sbi    = ll_i2sbi(inode);
-       __u64                   pos;
        int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
        int                     api32   = ll_need_32bit_api(sbi);
+       struct md_op_data       *op_data;
+       __u64                   pos;
        int                     rc;
-#ifdef HAVE_TOUCH_ATIME_1ARG
-       struct path             path;
-#endif
        ENTRY;
 
        if (lfd != NULL)
@@ -609,8 +313,8 @@ static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
        else
                pos = 0;
 
-       CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu "
-              " 32bit_api %d\n", inode->i_ino, inode->i_generation,
+       CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
+              "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
               inode, (unsigned long)pos, i_size_read(inode), api32);
 
        if (pos == MDS_DIR_END_OFF)
@@ -619,32 +323,61 @@ static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
                 */
                GOTO(out, rc = 0);
 
-       rc = ll_dir_read(inode, &pos, cookie, filldir);
-       if (lfd != NULL)
-               lfd->lfd_pos = pos;
-        if (pos == MDS_DIR_END_OFF) {
-                if (api32)
-                        filp->f_pos = LL_DIR_END_OFF_32BIT;
-                else
-                        filp->f_pos = LL_DIR_END_OFF;
-        } else {
-                if (api32 && hash64)
-                        filp->f_pos = pos >> 32;
-                else
-                        filp->f_pos = pos;
-        }
-       filp->f_version = inode->i_version;
-#ifdef HAVE_TOUCH_ATIME_1ARG
-#ifdef HAVE_F_PATH_MNT
-       path.mnt = filp->f_path.mnt;
+       op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
+                                    LUSTRE_OPC_ANY, inode);
+       if (IS_ERR(op_data))
+               GOTO(out, rc = PTR_ERR(op_data));
+
+       if (unlikely(op_data->op_mea1 != NULL)) {
+               /* This is only needed for striped dir to fill ..,
+                * see lmv_read_entry */
+               if (filp->f_dentry->d_parent != NULL &&
+                   filp->f_dentry->d_parent->d_inode != NULL) {
+                       __u64 ibits = MDS_INODELOCK_UPDATE;
+                       struct inode *parent =
+                               filp->f_dentry->d_parent->d_inode;
+
+                       if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
+                               op_data->op_fid3 = *ll_inode2fid(parent);
+               }
+
+               /* If it can not find in cache, do lookup .. on the master
+                * object */
+               if (fid_is_zero(&op_data->op_fid3)) {
+                       rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
+                       if (rc != 0) {
+                               ll_finish_md_op_data(op_data);
+                               RETURN(rc);
+                       }
+               }
+       }
+       op_data->op_max_pages = sbi->ll_md_brw_pages;
+#ifdef HAVE_DIR_CONTEXT
+       ctx->pos = pos;
+       rc = ll_dir_read(inode, &pos, op_data, ctx);
+       pos = ctx->pos;
 #else
-       path.mnt = filp->f_vfsmnt;
+       rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 #endif
-       path.dentry = filp->f_dentry;
-       touch_atime(&path);
+       if (lfd != NULL)
+               lfd->lfd_pos = pos;
+
+       if (pos == MDS_DIR_END_OFF) {
+               if (api32)
+                       pos = LL_DIR_END_OFF_32BIT;
+               else
+                       pos = LL_DIR_END_OFF;
+       } else {
+               if (api32 && hash64)
+                       pos = pos >> 32;
+       }
+#ifdef HAVE_DIR_CONTEXT
+       ctx->pos = pos;
 #else
-       touch_atime(filp->f_vfsmnt, filp->f_dentry);
+       filp->f_pos = pos;
 #endif
+       ll_finish_md_op_data(op_data);
+       filp->f_version = inode->i_version;
 
 out:
        if (!rc)
@@ -653,7 +386,7 @@ out:
        RETURN(rc);
 }
 
-int ll_send_mgc_param(struct obd_export *mgc, char *string)
+static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 {
         struct mgs_send_param *msp;
         int rc = 0;
@@ -662,7 +395,7 @@ int ll_send_mgc_param(struct obd_export *mgc, char *string)
         if (!msp)
                 return -ENOMEM;
 
-        strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
+       strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
                                 sizeof(struct mgs_send_param), msp, NULL);
         if (rc)
@@ -672,27 +405,50 @@ int ll_send_mgc_param(struct obd_export *mgc, char *string)
         return rc;
 }
 
-int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
-                       char *filename)
+/**
+ * Create striped directory with specified stripe(@lump)
+ *
+ * param[in]parent     the parent of the directory.
+ * param[in]lump       the specified stripes.
+ * param[in]dirname    the name of the directory.
+ * param[in]mode       the specified mode of the directory.
+ *
+ * retval              =0 if striped directory is being created successfully.
+ *                      <0 if the creation is failed.
+ */
+static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
+                              const char *dirname, umode_t mode)
 {
        struct ptlrpc_request *request = NULL;
        struct md_op_data *op_data;
-       struct ll_sb_info *sbi = ll_i2sbi(dir);
-       int mode;
+       struct ll_sb_info *sbi = ll_i2sbi(parent);
        int err;
-
        ENTRY;
 
-       mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
-       op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
-                                    strlen(filename), mode, LUSTRE_OPC_MKDIR,
+       if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
+               RETURN(-EINVAL);
+
+       CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
+              "stripe_offset %d, stripe_count: %u\n",
+              PFID(ll_inode2fid(parent)), parent, dirname,
+              (int)lump->lum_stripe_offset, lump->lum_stripe_count);
+
+       if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
+               lustre_swab_lmv_user_md(lump);
+
+       if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
+               mode &= ~current_umask();
+       mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
+       op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
+                                    strlen(dirname), mode, LUSTRE_OPC_MKDIR,
                                     lump);
        if (IS_ERR(op_data))
                GOTO(err_exit, err = PTR_ERR(op_data));
 
        op_data->op_cli_flags |= CLI_SET_MEA;
        err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
-                       current_fsuid(), current_fsgid(),
+                       from_kuid(&init_user_ns, current_fsuid()),
+                       from_kgid(&init_user_ns, current_fsgid()),
                        cfs_curproc_cap_pack(), 0, &request);
        ll_finish_md_op_data(op_data);
        if (err)
@@ -734,6 +490,13 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
                         lum_size = sizeof(struct lov_user_md_v3);
                         break;
                 }
+               case LMV_USER_MAGIC: {
+                       if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
+                               lustre_swab_lmv_user_md(
+                                       (struct lmv_user_md *)lump);
+                       lum_size = sizeof(struct lmv_user_md);
+                       break;
+               }
                 default: {
                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
                                         " %#08x != %#08x nor %#08x\n",
@@ -751,9 +514,6 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
         if (IS_ERR(op_data))
                 RETURN(PTR_ERR(op_data));
 
-       if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
-               op_data->op_cli_flags |= CLI_SET_MEA;
-
         /* swabbing is done in lov_setstripe() on server side */
         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
                         NULL, 0, &req, NULL);
@@ -808,74 +568,112 @@ end:
        RETURN(rc);
 }
 
-int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
-                     int *lmm_size, struct ptlrpc_request **request)
+/**
+ * This function will be used to get default LOV/LMV/Default LMV
+ * @valid will be used to indicate which stripe it will retrieve
+ *     OBD_MD_MEA              LMV stripe EA
+ *     OBD_MD_DEFAULT_MEA      Default LMV stripe EA
+ *     otherwise               Default LOV EA.
+ * Each time, it can only retrieve 1 stripe EA
+ **/
+int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
+                    struct ptlrpc_request **request, obd_valid valid)
 {
-        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 md_op_data *op_data;
+       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, lmm_size;
+       struct md_op_data *op_data;
+       ENTRY;
 
-        rc = ll_get_max_mdsize(sbi, &lmmsize);
-        if (rc)
-                RETURN(rc);
+       rc = ll_get_default_mdsize(sbi, &lmm_size);
+       if (rc)
+               RETURN(rc);
 
-        op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
-                                     0, lmmsize, LUSTRE_OPC_ANY,
-                                     NULL);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
+       op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
+                                    0, lmm_size, LUSTRE_OPC_ANY,
+                                    NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
 
-        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,
-                       inode->i_generation, rc);
-                GOTO(out, rc);
-        }
+       op_data->op_valid = 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 "
+                      DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
+               GOTO(out, rc);
+       }
 
-        body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-        LASSERT(body != NULL);
+       body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+       LASSERT(body != NULL);
 
-        lmmsize = body->eadatasize;
+       lmm_size = body->mbo_eadatasize;
 
-        if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
-            lmmsize == 0) {
-                GOTO(out, rc = -ENODATA);
-        }
+       if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
+           lmm_size == 0) {
+               GOTO(out, rc = -ENODATA);
+       }
 
-        lmm = req_capsule_server_sized_get(&req->rq_pill,
-                                           &RMF_MDT_MD, lmmsize);
-        LASSERT(lmm != NULL);
-
-        /*
-         * This is coming from the MDS, so is probably in
-         * little endian.  We convert it to host endian before
-         * passing it to userspace.
-         */
-        /* 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;
-        }
+       lmm = req_capsule_server_sized_get(&req->rq_pill,
+                                          &RMF_MDT_MD, lmm_size);
+       LASSERT(lmm != NULL);
+
+       /*
+        * This is coming from the MDS, so is probably in
+        * little endian.  We convert it to host endian before
+        * passing it to userspace.
+        */
+       /* 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;
+       case LMV_MAGIC_V1:
+               if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
+                       lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
+               break;
+       case LMV_USER_MAGIC:
+               if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
+                       lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
+               break;
+       default:
+               CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
+               rc = -EPROTO;
+       }
 out:
-        *lmmp = lmm;
-        *lmm_size = lmmsize;
-        *request = req;
-        return rc;
+       *plmm = lmm;
+       *plmm_size = lmm_size;
+       *request = req;
+       return rc;
+}
+
+int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
+{
+       struct md_op_data       *op_data;
+       int                     rc;
+       int                     mdt_index;
+       ENTRY;
+
+       OBD_ALLOC_PTR(op_data);
+       if (op_data == NULL)
+               RETURN(-ENOMEM);
+
+       op_data->op_flags |= MF_GET_MDT_IDX;
+       op_data->op_fid1 = *fid;
+       rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
+       mdt_index = op_data->op_mds;
+       OBD_FREE_PTR(op_data);
+       if (rc < 0)
+               RETURN(rc);
+
+       RETURN(mdt_index);
 }
 
 /*
@@ -883,25 +681,7 @@ out:
  */
 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 (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
-
-       op_data->op_flags |= MF_GET_MDT_IDX;
-        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;
+       return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 }
 
 /**
@@ -947,7 +727,7 @@ static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
                }
 
                /* Read current file data version */
-               rc = ll_data_version(inode, &data_version, 1);
+               rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
                iput(inode);
                if (rc != 0) {
                        CDEBUG(D_HSM, "Could not read file data version of "
@@ -1031,8 +811,7 @@ static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
                        GOTO(progress, rc = PTR_ERR(inode));
                }
 
-               rc = ll_data_version(inode, &data_version,
-                                    copy->hc_hai.hai_action == HSMA_ARCHIVE);
+               rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
                iput(inode);
                if (rc) {
                        CDEBUG(D_HSM, "Could not read file data version. "
@@ -1120,8 +899,10 @@ static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
                         RETURN(-EPERM);
                 break;
        case Q_GETQUOTA:
-               if (((type == USRQUOTA && current_euid() != id) ||
-                    (type == GRPQUOTA && !in_egroup_p(id))) &&
+               if (((type == USRQUOTA &&
+                     !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
+                    (type == GRPQUOTA &&
+                     !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
                    (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
                     sbi->ll_flags & LL_SBI_RMT_CLIENT))
                        RETURN(-EPERM);
@@ -1267,8 +1048,8 @@ static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
         int rc = 0;
         ENTRY;
 
-        CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
-               inode->i_ino, inode->i_generation, inode, cmd);
+       CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
+              PFID(ll_inode2fid(inode)), inode, cmd);
 
         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
@@ -1301,39 +1082,29 @@ static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                 return 0;
         }
         case IOC_MDC_LOOKUP: {
-                struct ptlrpc_request *request = NULL;
-                int namelen, len = 0;
-                char *buf = NULL;
-                char *filename;
-                struct md_op_data *op_data;
+               int namelen, len = 0;
+               char *buf = NULL;
+               char *filename;
 
-                rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
-                if (rc)
-                        RETURN(rc);
-                data = (void *)buf;
-
-                filename = data->ioc_inlbuf1;
-                namelen = strlen(filename);
-
-                if (namelen < 1) {
-                        CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
-                        GOTO(out_free, rc = -EINVAL);
-                }
+               rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
+               if (rc != 0)
+                       RETURN(rc);
+               data = (void *)buf;
 
-                op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
-                                             0, LUSTRE_OPC_ANY, NULL);
-                if (IS_ERR(op_data))
-                        GOTO(out_free, rc = PTR_ERR(op_data));
+               filename = data->ioc_inlbuf1;
+               namelen = strlen(filename);
+               if (namelen < 1) {
+                       CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
+                       GOTO(out_free, rc = -EINVAL);
+               }
 
-                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_free, rc);
-                }
-                ptlrpc_req_finished(request);
-                EXIT;
+               rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
+               if (rc < 0) {
+                       CERROR("%s: lookup %.*s failed: rc = %d\n",
+                              ll_get_fsname(inode->i_sb, NULL, 0), namelen,
+                              filename, rc);
+                       GOTO(out_free, rc);
+               }
 out_free:
                 obd_ioctl_freedata(buf, len);
                 return rc;
@@ -1344,6 +1115,7 @@ out_free:
                char            *filename;
                int              namelen = 0;
                int              lumlen = 0;
+               umode_t          mode;
                int              len;
                int              rc;
 
@@ -1373,16 +1145,33 @@ out_free:
                        GOTO(lmv_out_free, rc = -EINVAL);
                }
 
-               /**
-                * ll_dir_setdirstripe will be used to set dir stripe
-                *  mdc_create--->mdt_reint_create (with dirstripe)
-                */
-               rc = ll_dir_setdirstripe(inode, lum, filename);
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
+               mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
+#else
+               mode = data->ioc_type;
+#endif
+               rc = ll_dir_setdirstripe(inode, lum, filename, mode);
 lmv_out_free:
                obd_ioctl_freedata(buf, len);
                RETURN(rc);
 
        }
+       case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
+               struct lmv_user_md        lum;
+               struct lmv_user_md __user *ulump =
+                                       (struct lmv_user_md __user *)arg;
+               int                       rc;
+
+               if (copy_from_user(&lum, ulump, sizeof(lum)))
+                       RETURN(-EFAULT);
+
+               if (lum.lum_magic != LMV_USER_MAGIC)
+                       RETURN(-EINVAL);
+
+               rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
+
+               RETURN(rc);
+       }
         case LL_IOC_LOV_SETSTRIPE: {
                 struct lov_user_md_v3 lumv3;
                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
@@ -1412,42 +1201,87 @@ lmv_out_free:
                 RETURN(rc);
         }
        case LL_IOC_LMV_GETSTRIPE: {
-               struct lmv_user_md *lump = (struct lmv_user_md *)arg;
-               struct lmv_user_md lum;
-               struct lmv_user_md *tmp;
-               int lum_size;
-               int rc = 0;
-               int mdtindex;
-
-               if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
+               struct lmv_user_md __user *ulmv =
+                                       (struct lmv_user_md __user *)arg;
+               struct lmv_user_md      lum;
+               struct ptlrpc_request   *request = NULL;
+               union lmv_mds_md        *lmm = NULL;
+               int                     lmmsize;
+               obd_valid               valid = 0;
+               struct lmv_user_md      *tmp = NULL;
+               int                     mdt_index;
+               int                     lum_size;
+               int                     stripe_count;
+               int                     i;
+               int                     rc;
+
+               if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
                        RETURN(-EFAULT);
 
-               if (lum.lum_magic != LMV_MAGIC_V1)
+               /* lum_magic will indicate which stripe the ioctl will like
+                * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
+                * is for default LMV stripe */
+               if (lum.lum_magic == LMV_MAGIC_V1)
+                       valid |= OBD_MD_MEA;
+               else if (lum.lum_magic == LMV_USER_MAGIC)
+                       valid |= OBD_MD_DEFAULT_MEA;
+               else
                        RETURN(-EINVAL);
 
-               lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
+               rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
+                                     valid);
+               if (rc != 0)
+                       GOTO(finish_req, rc);
+
+               /* Get default LMV EA */
+               if (lum.lum_magic == LMV_USER_MAGIC) {
+                       if (rc != 0)
+                               GOTO(finish_req, rc);
+
+                       if (lmmsize > sizeof(*ulmv))
+                               GOTO(finish_req, rc = -EINVAL);
+
+                       if (copy_to_user(ulmv, lmm, lmmsize))
+                               GOTO(finish_req, rc = -EFAULT);
+
+                       GOTO(finish_req, rc);
+               }
+
+               stripe_count = lmv_mds_md_stripe_count_get(lmm);
+               lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
                OBD_ALLOC(tmp, lum_size);
                if (tmp == NULL)
-                       GOTO(free_lmv, rc = -ENOMEM);
-
-               memcpy(tmp, &lum, sizeof(lum));
-               tmp->lum_type = LMV_STRIPE_TYPE;
-               tmp->lum_stripe_count = 1;
-               mdtindex = ll_get_mdt_idx(inode);
-               if (mdtindex < 0)
-                       GOTO(free_lmv, rc = -ENOMEM);
-
-               tmp->lum_stripe_offset = mdtindex;
-               tmp->lum_objects[0].lum_mds = mdtindex;
-               memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
-                      sizeof(struct lu_fid));
-               if (copy_to_user((void *)arg, tmp, lum_size))
-                       GOTO(free_lmv, rc = -EFAULT);
-free_lmv:
-               if (tmp)
-                       OBD_FREE(tmp, lum_size);
-               RETURN(rc);
+                       GOTO(finish_req, rc = -ENOMEM);
+
+               mdt_index = ll_get_mdt_idx(inode);
+               if (mdt_index < 0)
+                       GOTO(out_tmp, rc = -ENOMEM);
+
+               tmp->lum_magic = LMV_MAGIC_V1;
+               tmp->lum_stripe_count = 0;
+               tmp->lum_stripe_offset = mdt_index;
+               for (i = 0; i < stripe_count; i++) {
+                       struct lu_fid   *fid;
+
+                       fid = &lmm->lmv_md_v1.lmv_stripe_fids[i];
+                       mdt_index = ll_get_mdt_idx_by_fid(sbi, fid);
+                       if (mdt_index < 0)
+                               GOTO(out_tmp, rc = mdt_index);
+
+                       tmp->lum_objects[i].lum_mds = mdt_index;
+                       tmp->lum_objects[i].lum_fid = *fid;
+                       tmp->lum_stripe_count++;
+               }
+
+               if (copy_to_user(ulmv, tmp, lum_size))
+                       GOTO(out_tmp, rc = -EFAULT);
+out_tmp:
+               OBD_FREE(tmp, lum_size);
+finish_req:
+               ptlrpc_req_finished(request);
+               return rc;
        }
+
        case LL_IOC_REMOVE_ENTRY: {
                char            *filename = NULL;
                int              namelen = 0;
@@ -1477,8 +1311,8 @@ out_rmdir:
        }
        case LL_IOC_LOV_SWAP_LAYOUTS:
                RETURN(-EPERM);
-        case LL_IOC_OBD_STATFS:
-                RETURN(ll_obd_statfs(inode, (void *)arg));
+       case IOC_OBD_STATFS:
+               RETURN(ll_obd_statfs(inode, (void *)arg));
         case LL_IOC_LOV_GETSTRIPE:
         case LL_IOC_MDC_GETINFO:
         case IOC_MDC_GETFILEINFO:
@@ -1498,9 +1332,10 @@ out_rmdir:
 
                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
                                                       &lmmsize, &request);
-                } else {
-                        rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
-                }
+               } else {
+                       rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
+                                             &request, 0);
+               }
 
                 if (request) {
                         body = req_capsule_server_get(&request->rq_pill,
@@ -1536,21 +1371,21 @@ out_rmdir:
                         struct lov_user_mds_data *lmdp;
                         lstat_t st = { 0 };
 
-                        st.st_dev     = inode->i_sb->s_dev;
-                        st.st_mode    = body->mode;
-                        st.st_nlink   = body->nlink;
-                        st.st_uid     = body->uid;
-                        st.st_gid     = body->gid;
-                        st.st_rdev    = body->rdev;
-                        st.st_size    = body->size;
-                       st.st_blksize = PAGE_CACHE_SIZE;
-                        st.st_blocks  = body->blocks;
-                        st.st_atime   = body->atime;
-                        st.st_mtime   = body->mtime;
-                        st.st_ctime   = body->ctime;
-                        st.st_ino     = inode->i_ino;
-
-                        lmdp = (struct lov_user_mds_data *)arg;
+                       st.st_dev       = inode->i_sb->s_dev;
+                       st.st_mode      = body->mbo_mode;
+                       st.st_nlink     = body->mbo_nlink;
+                       st.st_uid       = body->mbo_uid;
+                       st.st_gid       = body->mbo_gid;
+                       st.st_rdev      = body->mbo_rdev;
+                       st.st_size      = body->mbo_size;
+                       st.st_blksize   = PAGE_CACHE_SIZE;
+                       st.st_blocks    = body->mbo_blocks;
+                       st.st_atime     = body->mbo_atime;
+                       st.st_mtime     = body->mbo_mtime;
+                       st.st_ctime     = body->mbo_ctime;
+                       st.st_ino       = inode->i_ino;
+
+                       lmdp = (struct lov_user_mds_data *)arg;
                        if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
                                 GOTO(out_req, rc = -EFAULT);
                 }
@@ -1628,10 +1463,7 @@ out_rmdir:
                 OBD_FREE_LARGE(lmm, lmmsize);
                 return rc;
         }
-        case OBD_IOC_LLOG_CATINFO: {
-               RETURN(-EOPNOTSUPP);
-        }
-        case OBD_IOC_QUOTACHECK: {
+       case OBD_IOC_QUOTACHECK: {
                 struct obd_quotactl *oqctl;
                 int error = 0;
 
@@ -1656,7 +1488,7 @@ out_rmdir:
                 OBD_FREE_PTR(oqctl);
                 return error ?: rc;
         }
-        case OBD_IOC_POLL_QUOTACHECK: {
+       case OBD_IOC_POLL_QUOTACHECK: {
                 struct if_quotacheck *check;
 
                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
@@ -1744,7 +1576,7 @@ out_rmdir:
 #else
 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
-        case LL_IOC_QUOTACTL: {
+       case OBD_IOC_QUOTACTL: {
                 struct if_quotactl *qctl;
 
                 OBD_ALLOC_PTR(qctl);
@@ -1822,7 +1654,7 @@ out_rmdir:
                RETURN(ll_fid2path(inode, (void *)arg));
        case LL_IOC_HSM_REQUEST: {
                struct hsm_user_request *hur;
-               int                      totalsize;
+               ssize_t                  totalsize;
 
                OBD_ALLOC_PTR(hur);
                if (hur == NULL)
@@ -1837,9 +1669,11 @@ out_rmdir:
                /* Compute the whole struct size */
                totalsize = hur_len(hur);
                OBD_FREE_PTR(hur);
+               if (totalsize < 0)
+                       RETURN(-E2BIG);
 
-               /* Make sure the size is reasonable */
-               if (totalsize >= MDS_MAXREQSIZE)
+               /* Final size will be more than double totalsize */
+               if (totalsize >= MDS_MAXREQSIZE / 3)
                        RETURN(-E2BIG);
 
                OBD_ALLOC_LARGE(hur, totalsize);
@@ -1900,6 +1734,9 @@ out_rmdir:
                RETURN(rc);
        }
        case LL_IOC_HSM_CT_START:
+               if (!cfs_capable(CFS_CAP_SYS_ADMIN))
+                       RETURN(-EPERM);
+
                rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
                                    sizeof(struct lustre_kernelcomm));
                RETURN(rc);
@@ -1942,6 +1779,39 @@ out_rmdir:
                OBD_FREE_PTR(copy);
                RETURN(rc);
        }
+       case LL_IOC_MIGRATE: {
+               char            *buf = NULL;
+               const char      *filename;
+               int             namelen = 0;
+               int             len;
+               int             rc;
+               int             mdtidx;
+
+               rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
+               if (rc < 0)
+                       RETURN(rc);
+
+               data = (struct obd_ioctl_data *)buf;
+               if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
+                   data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
+                       GOTO(migrate_free, rc = -EINVAL);
+
+               filename = data->ioc_inlbuf1;
+               namelen = data->ioc_inllen1;
+               /* \0 is packed at the end of filename */
+               if (namelen < 1 || namelen != strlen(filename) + 1)
+                       GOTO(migrate_free, rc = -EINVAL);
+
+               if (data->ioc_inllen2 != sizeof(mdtidx))
+                       GOTO(migrate_free, rc = -EINVAL);
+               mdtidx = *(int *)data->ioc_inlbuf2;
+
+               rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
+migrate_free:
+               obd_ioctl_freedata(buf, len);
+
+               RETURN(rc);
+       }
        default:
                RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
                                     (void *)arg));
@@ -1999,24 +1869,28 @@ out:
         return ret;
 }
 
-int ll_dir_open(struct inode *inode, struct file *file)
+static int ll_dir_open(struct inode *inode, struct file *file)
 {
         ENTRY;
         RETURN(ll_file_open(inode, file));
 }
 
-int ll_dir_release(struct inode *inode, struct file *file)
+static int ll_dir_release(struct inode *inode, struct file *file)
 {
         ENTRY;
         RETURN(ll_file_release(inode, file));
 }
 
-struct file_operations ll_dir_operations = {
-        .llseek   = ll_dir_seek,
-        .open     = ll_dir_open,
-        .release  = ll_dir_release,
-        .read     = generic_read_dir,
-        .readdir  = ll_readdir,
-        .unlocked_ioctl   = ll_dir_ioctl,
-        .fsync    = ll_fsync,
+const struct file_operations ll_dir_operations = {
+       .llseek         = ll_dir_seek,
+       .open           = ll_dir_open,
+       .release        = ll_dir_release,
+       .read           = generic_read_dir,
+#ifdef HAVE_DIR_CONTEXT
+       .iterate        = ll_iterate,
+#else
+       .readdir        = ll_readdir,
+#endif
+       .unlocked_ioctl = ll_dir_ioctl,
+       .fsync          = ll_fsync,
 };