Whamcloud - gitweb
- landing b_fid.
[fs/lustre-release.git] / lustre / mds / mds_lmv.c
index 228cbfc..e4200a2 100644 (file)
@@ -44,7 +44,7 @@
  *   - error handling is totally missed
  */
 
-int mds_lmv_connect(struct obd_device *obd, char * lmv_name)
+int mds_lmv_connect(struct obd_device *obd, char *lmv_name)
 {
         struct mds_obd *mds = &obd->u.mds;
         struct lustre_handle conn = {0};
@@ -54,23 +54,30 @@ int mds_lmv_connect(struct obd_device *obd, char * lmv_name)
         if (IS_ERR(mds->mds_lmv_obd))
                 RETURN(PTR_ERR(mds->mds_lmv_obd));
 
-        if (mds->mds_lmv_obd)
+        if (mds->mds_lmv_connected)
                 RETURN(0);
 
+        down(&mds->mds_lmv_sem);
+        if (mds->mds_lmv_connected) {
+                up(&mds->mds_lmv_sem);
+                RETURN(0);
+        }
+
         mds->mds_lmv_obd = class_name2obd(lmv_name);
         if (!mds->mds_lmv_obd) {
                 CERROR("MDS cannot locate LMV %s\n",
                        lmv_name);
                 mds->mds_lmv_obd = ERR_PTR(-ENOTCONN);
-                RETURN(-ENOTCONN);
+                GOTO(err_last, rc = -ENOTCONN);
         }
 
-        rc = obd_connect(&conn, mds->mds_lmv_obd, &obd->obd_uuid);
+        rc = obd_connect(&conn, mds->mds_lmv_obd,
+                         &obd->obd_uuid, OBD_OPT_MDS_CONNECTION);
         if (rc) {
                 CERROR("MDS cannot connect to LMV %s (%d)\n",
                        lmv_name, rc);
                 mds->mds_lmv_obd = ERR_PTR(rc);
-                RETURN(rc);
+                GOTO(err_last, rc);
         }
         mds->mds_lmv_exp = class_conn2export(&conn);
         if (mds->mds_lmv_exp == NULL)
@@ -84,8 +91,8 @@ int mds_lmv_connect(struct obd_device *obd, char * lmv_name)
         }
 
         /* retrieve size of EA */
-        rc = obd_get_info(mds->mds_lmv_exp, strlen("mdsize"), "mdsize", 
-                          &valsize, &value);
+        rc = obd_get_info(mds->mds_lmv_exp, strlen("mdsize"),
+                          "mdsize", &valsize, &value);
         if (rc) 
                 GOTO(err_reg, rc);
 
@@ -93,8 +100,8 @@ int mds_lmv_connect(struct obd_device *obd, char * lmv_name)
                 mds->mds_max_mdsize = value;
 
         /* find our number in LMV cluster */
-        rc = obd_get_info(mds->mds_lmv_exp, strlen("mdsnum"), "mdsnum", 
-                          &valsize, &value);
+        rc = obd_get_info(mds->mds_lmv_exp, strlen("mdsnum"),
+                          "mdsnum", &valsize, &value);
         if (rc) 
                 GOTO(err_reg, rc);
         
@@ -104,7 +111,9 @@ int mds_lmv_connect(struct obd_device *obd, char * lmv_name)
                           "inter_mds", 0, NULL);
         if (rc)
                 GOTO(err_reg, rc);
-        
+
+        mds->mds_lmv_connected = 1;
+        up(&mds->mds_lmv_sem);
        RETURN(0);
 
 err_reg:
@@ -113,6 +122,8 @@ err_discon:
         obd_disconnect(mds->mds_lmv_exp, 0);
         mds->mds_lmv_exp = NULL;
         mds->mds_lmv_obd = ERR_PTR(rc);
+err_last:
+        up(&mds->mds_lmv_sem);
         RETURN(rc);
 }
 
@@ -122,9 +133,11 @@ int mds_lmv_postsetup(struct obd_device *obd)
         int rc = 0;
         ENTRY;
 
-        if (mds->mds_lmv_exp)
-                rc = obd_init_ea_size(mds->mds_lmv_exp, mds->mds_max_mdsize,
+        if (mds->mds_lmv_exp) {
+                rc = obd_init_ea_size(mds->mds_lmv_exp,
+                                      mds->mds_max_mdsize,
                                       mds->mds_max_cookiesize);
+        }
         
         RETURN(rc);
 }
@@ -135,20 +148,52 @@ int mds_lmv_disconnect(struct obd_device *obd, int flags)
         int rc = 0;
         ENTRY;
 
+        if (!mds->mds_lmv_connected)
+                RETURN(0);
+
+        down(&mds->mds_lmv_sem);
         if (!IS_ERR(mds->mds_lmv_obd) && mds->mds_lmv_exp != NULL) {
+                LASSERT(mds->mds_lmv_connected);
+                
                 obd_register_observer(mds->mds_lmv_obd, NULL);
 
-                /* if obd_disconnect fails (probably because the export was
-                 * disconnected by class_disconnect_exports) then we just need
-                 * to drop our ref. */
+                if (flags & OBD_OPT_FORCE) {
+                        struct obd_device *lmv_obd;
+                        struct obd_ioctl_data ioc_data = { 0 };
+                        
+                        lmv_obd = class_exp2obd(mds->mds_lmv_exp);
+                        if (lmv_obd == NULL)
+                                GOTO(out, rc = 0);
+
+                        /* 
+                         * making disconnecting lmv stuff do not send anything
+                         * to all remote MDSs from LMV. This is needed to
+                         * prevent possible hanging with endless recovery, when
+                         * MDS sends disconnect to already disconnected
+                         * target. Probably this is wrong, but client does the
+                         * same in --force mode and I do not see why can't we do
+                         * it here. --umka.
+                         */
+                        lmv_obd->obd_no_recov = 1;
+                        obd_iocontrol(IOC_OSC_SET_ACTIVE, mds->mds_lmv_exp,
+                                      sizeof(ioc_data), &ioc_data, NULL);
+                }
+
+                /*
+                 * if obd_disconnect() fails (probably because the export was
+                 * disconnected by class_disconnect_exports()) then we just need
+                 * to drop our ref.
+                 */
+                mds->mds_lmv_connected = 0;
                 rc = obd_disconnect(mds->mds_lmv_exp, flags);
                 if (rc)
                         class_export_put(mds->mds_lmv_exp);
-                
+
+        out:
                 mds->mds_lmv_exp = NULL;
                 mds->mds_lmv_obd = NULL;
         }
-
+        up(&mds->mds_lmv_sem);
         RETURN(rc);
 }
 
@@ -161,6 +206,8 @@ int mds_get_lmv_attr(struct obd_device *obd, struct inode *inode,
 
        if (!mds->mds_lmv_obd)
                RETURN(0);
+        if (!S_ISDIR(inode->i_mode))
+                RETURN(0);
 
        /* first calculate mea size */
         *mea_size = obd_alloc_diskmd(mds->mds_lmv_exp,
@@ -168,16 +215,14 @@ int mds_get_lmv_attr(struct obd_device *obd, struct inode *inode,
         if (*mea_size < 0 || *mea == NULL)
                 return *mea_size < 0 ? *mea_size : -EINVAL;
 
-       down(&inode->i_sem);
-       rc = fsfilt_get_md(obd, inode, *mea, *mea_size);
-       up(&inode->i_sem);
+        rc = mds_get_md(obd, inode, *mea, mea_size, 1);
 
        if (rc <= 0) {
                OBD_FREE(*mea, *mea_size);
                *mea = NULL;
        } else
                 rc = 0;
-                        
+
        RETURN(rc);
 }
 
@@ -186,12 +231,13 @@ struct dir_entry {
         __u16   mds;
         __u32   ino;
         __u32   generation;
+        __u32   fid;
         char    name[0];
 };
 
 #define DIR_PAD                        4
 #define DIR_ROUND              (DIR_PAD - 1)
-#define DIR_REC_LEN(name_len)  (((name_len) + 12 + DIR_ROUND) & ~DIR_ROUND)
+#define DIR_REC_LEN(name_len)  (((name_len) + 16 + DIR_ROUND) & ~DIR_ROUND)
 
 /* this struct holds dir entries for particular MDS to be flushed */
 struct dir_cache {
@@ -224,7 +270,7 @@ static int dc_new_page_to_cache(struct dir_cache * dirc)
         page = alloc_page(GFP_KERNEL);
         if (page == NULL)
                 return -ENOMEM;
-        list_add_tail(&page->list, &dirc->list);
+        list_add_tail(&page->lru, &dirc->list);
         dirc->cur = page_address(page);
         dirc->free = PAGE_SIZE;
         return 0;
@@ -235,26 +281,43 @@ static int retrieve_generation_numbers(struct dirsplit_control *dc, void *buf)
         struct mds_obd *mds = &dc->obd->u.mds;
         struct dir_entry *de;
         struct dentry *dentry;
-        char * end;
+        char *end;
         
         end = buf + PAGE_SIZE;
         de = (struct dir_entry *) buf;
         while ((char *) de < end && de->namelen) {
                 /* lookup an inode */
                 LASSERT(de->namelen <= 255);
-                dentry = ll_lookup_one_len(de->name, dc->dentry, de->namelen);
+                dentry = ll_lookup_one_len(de->name, dc->dentry, 
+                                           de->namelen);
                 if (IS_ERR(dentry)) {
                         CERROR("can't lookup %*s: %d\n", de->namelen,
                                de->name, (int) PTR_ERR(dentry));
                         goto next;
                 }
                 if (dentry->d_inode != NULL) {
+                        int rc;
+                        struct lustre_id sid;
+
+                        down(&dentry->d_inode->i_sem);
+                        rc = mds_read_inode_sid(dc->obd,
+                                                dentry->d_inode, &sid);
+                        up(&dentry->d_inode->i_sem);
+                        if (rc) {
+                                CERROR("Can't read inode self id, "
+                                       "inode %lu, rc %d\n",
+                                       dentry->d_inode->i_ino, rc);
+                                goto next;
+                        }
+
+                        de->fid = id_fid(&sid);
                         de->mds = mds->mds_num;
                         de->ino = dentry->d_inode->i_ino;
                         de->generation = dentry->d_inode->i_generation;
                 } else if (dentry->d_flags & DCACHE_CROSS_REF) {
-                        de->mds = dentry->d_mdsnum;
+                        de->fid = dentry->d_fid;
                         de->ino = dentry->d_inum;
+                        de->mds = dentry->d_mdsnum;
                         de->generation = dentry->d_generation;
                 } else {
                         CERROR("can't lookup %*s\n", de->namelen, de->name);
@@ -287,13 +350,13 @@ static int flush_buffer_onto_mds(struct dirsplit_control *dc, int mdsnum)
         list_for_each_safe(cur, tmp, &ca->list) {
                 struct page *page;
 
-                page = list_entry(cur, struct page, list);
+                page = list_entry(cur, struct page, lru);
                 LASSERT(page != NULL);
 
                 retrieve_generation_numbers(dc, page_address(page));
 
                 ca->brwc.pg = page;
-                ca->brwc.off = 0;
+                ca->brwc.disk_offset = ca->brwc.page_offset = 0;
                 ca->brwc.count = PAGE_SIZE;
                 ca->brwc.flag = 0;
                 ca->oa.o_mds = mdsnum;
@@ -320,7 +383,7 @@ static int remove_entries_from_orig_dir(struct dirsplit_control *dc, int mdsnum)
 
         ca = dc->cache + mdsnum;
         list_for_each_safe(cur, tmp, &ca->list) {
-                page = list_entry(cur, struct page, list);
+                page = list_entry(cur, struct page, lru);
                 buf = page_address(page);
                 end = buf + PAGE_SIZE;
 
@@ -329,11 +392,11 @@ static int remove_entries_from_orig_dir(struct dirsplit_control *dc, int mdsnum)
                         /* lookup an inode */
                         LASSERT(de->namelen <= 255);
 
-                        dentry = ll_lookup_one_len(de->name, dc->dentry,
+                        dentry = ll_lookup_one_len(de->name, dc->dentry, 
                                                    de->namelen);
                         if (IS_ERR(dentry)) {
                                 CERROR("can't lookup %*s: %d\n", de->namelen,
-                                                de->name, (int) PTR_ERR(dentry));
+                                       de->name, (int) PTR_ERR(dentry));
                                 goto next;
                         }
                         rc = fsfilt_del_dir_entry(dc->obd, dentry);
@@ -346,8 +409,8 @@ next:
         RETURN(0);
 }
 
-static int filldir(void * __buf, const char * name, int namlen, loff_t offset,
-                  ino_t ino, unsigned int d_type)
+static int filldir(void * __buf, const char * name, int namlen,
+                   loff_t offset, ino_t ino, unsigned int d_type)
 {
         struct dirsplit_control *dc = __buf;
         struct mds_obd *mds = &dc->obd->u.mds;
@@ -357,8 +420,8 @@ static int filldir(void * __buf, const char * name, int namlen, loff_t offset,
         char *n;
         ENTRY;
 
-        if (name[0] == '.' && (namlen == 1 ||
-                                (namlen == 2 && name[1] == '.'))) {
+        if (name[0] == '.' &&
+            (namlen == 1 || (namlen == 2 && name[1] == '.'))) {
                 /* skip special entries */
                 RETURN(0);
         }
@@ -401,7 +464,7 @@ static int filldir(void * __buf, const char * name, int namlen, loff_t offset,
 }
 
 int scan_and_distribute(struct obd_device *obd, struct dentry *dentry,
-                                struct mea *mea)
+                        struct mea *mea)
 {
         struct inode *dir = dentry->d_inode;
         struct dirsplit_control dc;
@@ -413,7 +476,9 @@ int scan_and_distribute(struct obd_device *obd, struct dentry *dentry,
         OBD_ALLOC(file_name, nlen);
         if (!file_name)
                 RETURN(-ENOMEM);
-        i = sprintf(file_name, "__iopen__/0x%lx", dentry->d_inode->i_ino);
+        
+        i = sprintf(file_name, "__iopen__/0x%lx",
+                    dentry->d_inode->i_ino);
 
         file = filp_open(file_name, O_RDONLY, 0);
         if (IS_ERR(file)) {
@@ -464,8 +529,8 @@ cleanup:
                         continue;
                 list_for_each_safe(cur, tmp, &dc.cache[i].list) {
                         struct page *page;
-                        page = list_entry(cur, struct page, list);
-                        list_del(&page->list);
+                        page = list_entry(cur, struct page, lru);
+                        list_del(&page->lru);
                         __free_page(page);
                 }
         }
@@ -475,7 +540,8 @@ cleanup:
         RETURN(err);
 }
 
-#define MAX_DIR_SIZE    (64 * 1024)
+#define MAX_DIR_SIZE      (64 * 1024)
+#define I_NON_SPLITTABLE  (256)
 
 int mds_splitting_expected(struct obd_device *obd, struct dentry *dentry)
 {
@@ -485,7 +551,7 @@ int mds_splitting_expected(struct obd_device *obd, struct dentry *dentry)
 
        /* clustered MD ? */
        if (!mds->mds_lmv_obd)
-               RETURN(MDS_NO_SPLITTABLE);
+               return MDS_NO_SPLITTABLE;
 
         /* inode exist? */
         if (dentry->d_inode == NULL)
@@ -495,8 +561,12 @@ int mds_splitting_expected(struct obd_device *obd, struct dentry *dentry)
         if (!S_ISDIR(dentry->d_inode->i_mode))
                 return MDS_NO_SPLITTABLE;
 
+        /* already splittied or slave directory (part of splitted dir) */
+        if (dentry->d_inode->i_flags & I_NON_SPLITTABLE)
+                return MDS_NO_SPLITTABLE;
+
         /* don't split root directory */
-        if (dentry->d_inode->i_ino == mds->mds_rootfid.id)
+        if (dentry->d_inode->i_ino == id_ino(&mds->mds_rootid))
                 return MDS_NO_SPLITTABLE;
 
         /* large enough to be splitted? */
@@ -507,45 +577,49 @@ int mds_splitting_expected(struct obd_device *obd, struct dentry *dentry)
         if (mea) {
                 /* already splitted or slave object: shouldn't be splitted */
                 rc = MDS_NO_SPLITTABLE;
+                /* mark to skip subsequent checks */
+                dentry->d_inode->i_flags |= I_NON_SPLITTABLE;
+                OBD_FREE(mea, size);
         } else {
                 /* may be splitted */
                 rc = MDS_EXPECT_SPLIT;
         }
 
-        if (mea)
-                OBD_FREE(mea, size);
-        RETURN(rc);
+        return rc;
 }
 
 /*
  * must not be called on already splitted directories.
  */
 int mds_try_to_split_dir(struct obd_device *obd, struct dentry *dentry,
-                         struct mea **mea, int nstripes)
+                         struct mea **mea, int nstripes, int update_mode)
 {
         struct inode *dir = dentry->d_inode;
         struct mds_obd *mds = &obd->u.mds;
         struct mea *tmea = NULL;
         struct obdo *oa = NULL;
        int rc, mea_size = 0;
+        struct lustre_id id;
        void *handle;
        ENTRY;
 
+        if (update_mode != LCK_EX)
+                return 0;
+        
         /* TODO: optimization possible - we already may have mea here */
         rc = mds_splitting_expected(obd, dentry);
         if (rc == MDS_NO_SPLITTABLE)
                 return 0;
         if (rc == MDS_NO_SPLIT_EXPECTED && nstripes == 0)
                 return 0;
+        if (nstripes && nstripes == 1)
+                return 0;
         
         LASSERT(mea == NULL || *mea == NULL);
 
         CDEBUG(D_OTHER, "%s: split directory %u/%lu/%lu\n",
                obd->obd_name, mds->mds_num, dir->i_ino,
                (unsigned long) dir->i_generation);
-        CDEBUG(D_ERROR, "%s: split directory %u/%lu/%lu\n",
-               obd->obd_name, mds->mds_num, dir->i_ino,
-               (unsigned long) dir->i_generation);
 
         if (mea == NULL)
                 mea = &tmea;
@@ -561,7 +635,7 @@ int mds_try_to_split_dir(struct obd_device *obd, struct dentry *dentry,
                 RETURN(rc);
         }
         if (*mea == NULL)
-                RETURN(-EINVAL);
+                RETURN(-ENOMEM);
 
         (*mea)->mea_count = nstripes;
        
@@ -571,7 +645,6 @@ int mds_try_to_split_dir(struct obd_device *obd, struct dentry *dentry,
         if (!oa)
                 RETURN(-ENOMEM);
 
-       oa->o_id = dir->i_ino;
        oa->o_generation = dir->i_generation;
 
         obdo_from_inode(oa, dir, OBD_MD_FLTYPE | OBD_MD_FLATIME |
@@ -581,15 +654,29 @@ int mds_try_to_split_dir(struct obd_device *obd, struct dentry *dentry,
         oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
         oa->o_valid |= OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
         oa->o_mode = dir->i_mode;
+       oa->o_id = dir->i_ino;
+
+        down(&dir->i_sem);
+        rc = mds_read_inode_sid(obd, dir, &id);
+        up(&dir->i_sem);
+        if (rc) {
+                CERROR("Can't read inode self id, inode %lu, "
+                       "rc %d.\n", dir->i_ino, rc);
+                GOTO(err_oa, rc);
+        }
+        oa->o_fid = id_fid(&id);
 
         CDEBUG(D_OTHER, "%s: create subdirs with mode %o, uid %u, gid %u\n",
-                        obd->obd_name, dir->i_mode, dir->i_uid, dir->i_gid);
+               obd->obd_name, dir->i_mode, dir->i_uid, dir->i_gid);
                         
         rc = obd_create(mds->mds_lmv_exp, oa,
                         (struct lov_stripe_md **)mea, NULL);
-        if (rc)
+        if (rc) {
+                CERROR("Can't create remote inode, rc = %d\n", rc);
                 GOTO(err_oa, rc);
+        }
 
+        LASSERT(id_fid(&(*mea)->mea_ids[0]));
         CDEBUG(D_OTHER, "%d dirobjects created\n", (int)(*mea)->mea_count);
 
        /* 2) update dir attribute */
@@ -601,7 +688,7 @@ int mds_try_to_split_dir(struct obd_device *obd, struct dentry *dentry,
                 CERROR("fsfilt_start() failed: %d\n", (int) PTR_ERR(handle));
                 GOTO(err_oa, rc = PTR_ERR(handle));
         }
-        
+
        rc = fsfilt_set_md(obd, dir, handle, *mea, mea_size);
         if (rc) {
                 up(&dir->i_sem);
@@ -650,9 +737,9 @@ static int filter_start_page_write(struct inode *inode,
         return 0;
 }
 
-struct dentry *filter_fid2dentry(struct obd_device *obd,
-                                 struct dentry *dir_dentry,
-                                 obd_gr group, obd_id id);
+struct dentry *filter_id2dentry(struct obd_device *obd,
+                                struct dentry *dir_dentry,
+                                obd_gr group, obd_id id);
 
 int mds_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
                 int objcount, struct obd_ioobj *obj,
@@ -660,25 +747,28 @@ int mds_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
                 struct niobuf_local *res,
                 struct obd_trans_info *oti)
 {
-        struct mds_obd *mds = &exp->exp_obd->u.mds;
         struct niobuf_remote *rnb;
         struct niobuf_local *lnb = NULL;
         int rc = 0, i, tot_bytes = 0;
         unsigned long now = jiffies;
         struct dentry *dentry;
-        struct ll_fid fid;
+        struct lustre_id id;
         ENTRY;
         LASSERT(objcount == 1);
         LASSERT(obj->ioo_bufcnt > 0);
 
         memset(res, 0, niocount * sizeof(*res));
 
-        fid.id = obj->ioo_id;
-        fid.generation = obj->ioo_gr;
-        dentry = mds_fid2dentry(mds, &fid, NULL);
+        id_fid(&id) = 0;
+        id_group(&id) = 0;
+        id_ino(&id) = obj->ioo_id;
+        id_gen(&id) = obj->ioo_gr;
+        
+        dentry = mds_id2dentry(exp->exp_obd, &id, NULL);
         if (IS_ERR(dentry)) {
                 CERROR("can't get dentry for "LPU64"/%u: %d\n",
-                       fid.id, fid.generation, (int) PTR_ERR(dentry));
+                       id.li_stc.u.e3s.l3s_ino, id.li_stc.u.e3s.l3s_gen,
+                       (int)PTR_ERR(dentry));
                 GOTO(cleanup, rc = (int) PTR_ERR(dentry));
         }
 
@@ -754,12 +844,12 @@ int mds_commitrw(int cmd, struct obd_export *exp, struct obdo *oa,
                 while ((char *) de < end && de->namelen) {
                         err = fsfilt_add_dir_entry(obd, res->dentry, de->name,
                                                    de->namelen, de->ino,
-                                                   de->generation, de->mds);
+                                                   de->generation, de->mds,
+                                                   de->fid);
                         if (err) {
                                 CERROR("can't add dir entry %*s->%u/%u/%u"
-                                       " to %lu/%u: %d\n",
-                                       de->namelen, de->name,
-                                       de->mds, (unsigned) de->ino,
+                                       " to %lu/%u: %d\n", de->namelen,
+                                       de->name, de->mds, (unsigned)de->ino,
                                        (unsigned) de->generation,
                                        res->dentry->d_inode->i_ino,
                                        res->dentry->d_inode->i_generation,
@@ -792,7 +882,7 @@ int mds_choose_mdsnum(struct obd_device *obd, const char *name, int len, int fla
                 i = mds->mds_num;
         } else if (mds->mds_lmv_exp) {
                 lmv = &mds->mds_lmv_exp->exp_obd->u.lmv;
-                i = raw_name2idx(lmv->desc.ld_tgt_count, name, len);
+                i = raw_name2idx(MEA_MAGIC_LAST_CHAR, lmv->desc.ld_tgt_count, name, len);
         }
         RETURN(i);
 }
@@ -805,6 +895,8 @@ int mds_lock_slave_objs(struct obd_device *obd, struct dentry *dentry,
         struct lookup_intent it;
         struct mea *mea = NULL;
         int mea_size, rc;
+        int handle_size;
+        ENTRY;
 
         LASSERT(rlockh != NULL);
         LASSERT(dentry != NULL);
@@ -812,38 +904,44 @@ int mds_lock_slave_objs(struct obd_device *obd, struct dentry *dentry,
 
        /* clustered MD ? */
        if (!mds->mds_lmv_obd)
-               return 0;
+               RETURN(0);
 
         /* a dir can be splitted only */
         if (!S_ISDIR(dentry->d_inode->i_mode))
-                return 0;
+                RETURN(0);
 
-        rc = mds_get_lmv_attr(obd, dentry->d_inode, &mea, &mea_size);
+        rc = mds_get_lmv_attr(obd, dentry->d_inode,
+                              &mea, &mea_size);
         if (rc)
-                return rc;
+                RETURN(rc);
 
         if (mea == NULL)
-                return 0;
-        if (mea->mea_count == 0) {
+                RETURN(0);
+        
+        if (mea->mea_count == 0)
                 /* this is slave object */
                 GOTO(cleanup, rc = 0);
-        }
                 
-        CDEBUG(D_OTHER, "%s: lock slaves for %lu/%lu\n", obd->obd_name,
-               (unsigned long) dentry->d_inode->i_ino,
-               (unsigned long) dentry->d_inode->i_generation);
+        CDEBUG(D_OTHER, "%s: lock slaves for %lu/%lu\n",
+               obd->obd_name, (unsigned long)dentry->d_inode->i_ino,
+               (unsigned long)dentry->d_inode->i_generation);
 
-        OBD_ALLOC(*rlockh, sizeof(struct lustre_handle) * mea->mea_count);
+        handle_size = sizeof(struct lustre_handle) *
+                mea->mea_count;
+        
+        OBD_ALLOC(*rlockh, handle_size);
         if (*rlockh == NULL)
                 GOTO(cleanup, rc = -ENOMEM);
-        memset(*rlockh, 0, sizeof(struct lustre_handle) * mea->mea_count);
 
+        memset(*rlockh, 0, handle_size);
         memset(&op_data, 0, sizeof(op_data));
+
         op_data.mea1 = mea;
         it.it_op = IT_UNLINK;
-        rc = md_enqueue(mds->mds_lmv_exp, LDLM_IBITS, &it, LCK_EX, &op_data,
-                        *rlockh, NULL, 0, ldlm_completion_ast, mds_blocking_ast,
-                        NULL);
+
+        rc = md_enqueue(mds->mds_lmv_exp, LDLM_IBITS, &it, LCK_EX,
+                        &op_data, *rlockh, NULL, 0, ldlm_completion_ast,
+                        mds_blocking_ast, NULL);
 cleanup:
         OBD_FREE(mea, mea_size);
         RETURN(rc);
@@ -950,6 +1048,7 @@ noempty:
         return -ENOTEMPTY;
 }
 
+/* checks if passed dentry points to empty dir. */
 int mds_is_dir_empty(struct obd_device *obd, struct dentry *dentry)
 {
         struct ide_tracking it;
@@ -964,7 +1063,10 @@ int mds_is_dir_empty(struct obd_device *obd, struct dentry *dentry)
         OBD_ALLOC(file_name, nlen);
         if (!file_name)
                 RETURN(-ENOMEM);
-        i = sprintf(file_name, "__iopen__/0x%lx", dentry->d_inode->i_ino);
+        
+        LASSERT(dentry->d_inode != NULL);
+        i = sprintf(file_name, "__iopen__/0x%lx",
+                    dentry->d_inode->i_ino);
 
         file = filp_open(file_name, O_RDONLY, 0);
         if (IS_ERR(file)) {
@@ -993,23 +1095,31 @@ int mds_lock_and_check_slave(int offset, struct ptlrpc_request *req,
         struct dentry *dentry = NULL;
         struct lvfs_run_ctxt saved;
         int cleanup_phase = 0;
+        struct mds_req_sec_desc *rsd;
         struct mds_body *body;
         struct lvfs_ucred uc;
         int rc, update_mode;
         ENTRY;
 
+        rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
+        if (!rsd) {
+                CERROR("Can't unpack security desc\n");
+                GOTO(cleanup, rc = -EFAULT);
+        }
+        mds_squash_root(&obd->u.mds, rsd, &req->rq_peer.peer_id.nid); 
+
         body = lustre_swab_reqbuf(req, offset, sizeof(*body),
                                   lustre_swab_mds_body);
         if (body == NULL) {
                 CERROR("Can't swab mds_body\n");
                 GOTO(cleanup, rc = -EFAULT);
         }
-        CDEBUG(D_OTHER, "%s: check slave %lu/%lu\n", obd->obd_name,
-               (unsigned long) body->fid1.id,
-               (unsigned long) body->fid1.generation);
-        dentry = mds_fid2locked_dentry(obd, &body->fid1, NULL, LCK_EX, lockh,
-                                       &update_mode, NULL, 0,
-                                       MDS_INODELOCK_UPDATE);
+        CDEBUG(D_OTHER, "%s: check slave "DLID4"\n", obd->obd_name,
+               OLID4(&body->id1));
+        
+        dentry = mds_id2locked_dentry(obd, &body->id1, NULL, LCK_EX,
+                                      lockh, &update_mode, NULL, 0,
+                                      MDS_INODELOCK_UPDATE);
         if (IS_ERR(dentry)) {
                 CERROR("can't find inode: %d\n", (int) PTR_ERR(dentry));
                 GOTO(cleanup, rc = PTR_ERR(dentry));
@@ -1018,11 +1128,12 @@ int mds_lock_and_check_slave(int offset, struct ptlrpc_request *req,
 
         LASSERT(S_ISDIR(dentry->d_inode->i_mode));
 
-        uc.luc_fsuid = body->fsuid;
-        uc.luc_fsgid = body->fsgid;
-        uc.luc_cap = body->capability;
-        uc.luc_suppgid1 = body->suppgid;
-        uc.luc_suppgid2 = -1;
+        rc = mds_init_ucred(&uc, rsd);
+        if (rc) {
+                CERROR("can't init ucred\n");
+                GOTO(cleanup, rc);
+        }
+
         push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
 
         rc = 0;
@@ -1036,9 +1147,75 @@ cleanup:
                         ldlm_lock_decref(lockh, LCK_EX);
                 l_dput(dentry);
                 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
+                mds_exit_ucred(&uc);
         default:
                 break;
         }
         RETURN(rc);
 }
 
+int mds_convert_mea_ea(struct obd_device *obd, struct inode *inode,
+                       struct lov_mds_md *lmm, int lmmsize)
+{
+        int i, rc, err, size;
+        struct mea_old *old;
+        struct mea *mea;
+        struct mea *new;
+        void *handle;
+        ENTRY;
+
+        mea = (struct mea *) lmm;
+        if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
+                mea->mea_magic == MEA_MAGIC_ALL_CHARS)
+                RETURN(0);
+
+        old = (struct mea_old *) lmm;
+        
+        rc = sizeof(struct lustre_id) * old->mea_count + 
+                sizeof(struct mea_old);
+        
+        if (old->mea_count > 256 || old->mea_master > 256 || lmmsize < rc
+                        || old->mea_master > old->mea_count) {
+                CWARN("unknown MEA format, dont convert it\n");
+                CWARN("  count %u, master %u, size %u\n",
+                      old->mea_count, old->mea_master, rc);
+                RETURN(0);
+        }
+                
+        CWARN("converting MEA EA on %lu/%u from V0 to V1 (%u/%u)\n",
+              inode->i_ino, inode->i_generation, old->mea_count, 
+              old->mea_master);
+
+        size = sizeof(struct lustre_id) * old->mea_count + 
+                sizeof(struct mea);
+        
+        OBD_ALLOC(new, size);
+        if (new == NULL)
+                RETURN(-ENOMEM);
+
+        new->mea_magic = MEA_MAGIC_LAST_CHAR;
+        new->mea_count = old->mea_count;
+        new->mea_master = old->mea_master;
+        for (i = 0; i < new->mea_count; i++)
+                new->mea_ids[i] = old->mea_ids[i];
+
+        handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
+        if (IS_ERR(handle)) {
+                rc = PTR_ERR(handle);
+                GOTO(conv_free, rc);
+        }
+
+        rc = fsfilt_set_md(obd, inode, handle, (struct lov_mds_md *) new, size);
+        if (rc > lmmsize)
+                size = lmmsize;
+        memcpy(lmm, new, size);
+
+        err = fsfilt_commit(obd, obd->u.mds.mds_sb, inode, handle, 0);
+        if (!rc)
+                rc = err ? err : size;
+        GOTO(conv_free, rc);
+conv_free:
+        OBD_FREE(new, size);
+        return rc;
+}
+