+static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
+{
+ struct mdd_device *mdd = (struct mdd_device *)data;
+ struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
+
+ LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
+ LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC);
+
+ CDEBUG(D_INFO,
+ "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n",
+ hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index,
+ rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name,
+ llh->lgh_id.lgl_oid);
+
+ mdd->mdd_cl.mc_index = rec->cr.cr_index;
+ return LLOG_PROC_BREAK;
+}
+
+static int changelog_user_init_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
+{
+ struct mdd_device *mdd = (struct mdd_device *)data;
+ struct llog_changelog_user_rec *rec =
+ (struct llog_changelog_user_rec *)hdr;
+
+ LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
+ LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC);
+
+ CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64
+ " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index,
+ rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid);
+
+ spin_lock(&mdd->mdd_cl.mc_user_lock);
+ mdd->mdd_cl.mc_lastuser = rec->cur_id;
+ spin_unlock(&mdd->mdd_cl.mc_user_lock);
+
+ return LLOG_PROC_BREAK;
+}
+
+static int llog_changelog_cancel_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
+{
+ struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
+ struct llog_cookie cookie;
+ long long endrec = *(long long *)data;
+ int rc;
+
+ ENTRY;
+
+ /* This is always a (sub)log, not the catalog */
+ LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN);
+
+ if (rec->cr.cr_index > endrec)
+ /* records are in order, so we're done */
+ RETURN(LLOG_PROC_BREAK);
+
+ cookie.lgc_lgl = llh->lgh_id;
+ cookie.lgc_index = hdr->lrh_index;
+
+ /* cancel them one at a time. I suppose we could store up the cookies
+ * and cancel them all at once; probably more efficient, but this is
+ * done as a user call, so who cares... */
+ rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1,
+ &cookie);
+ RETURN(rc < 0 ? rc : 0);
+}
+
+static int llog_changelog_cancel(const struct lu_env *env,
+ struct llog_ctxt *ctxt,
+ struct lov_stripe_md *lsm, int count,
+ struct llog_cookie *cookies, int flags)
+{
+ struct llog_handle *cathandle = ctxt->loc_handle;
+ int rc;
+
+ ENTRY;
+
+ /* This should only be called with the catalog handle */
+ LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT);
+
+ rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb,
+ (void *)cookies, 0, 0);
+ if (rc >= 0)
+ /* 0 or 1 means we're done */
+ rc = 0;
+ else
+ CERROR("%s: cancel idx %u of catalog "LPX64" rc=%d\n",
+ ctxt->loc_obd->obd_name, cathandle->lgh_last_idx,
+ cathandle->lgh_id.lgl_oid, rc);
+
+ RETURN(rc);
+}
+
+static struct llog_operations changelog_orig_logops;
+
+int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on);
+
+static int mdd_changelog_llog_init(const struct lu_env *env,
+ struct mdd_device *mdd)
+{
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_ctxt *ctxt = NULL, *uctxt = NULL;
+ struct lu_fid rfid;
+ int rc;
+
+ OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
+ obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom;
+ rc = dt_root_get(env, mdd->mdd_bottom, &rfid);
+ if (rc)
+ RETURN(-ENODEV);
+
+ changelog_orig_logops = llog_osd_ops;
+ changelog_orig_logops.lop_cancel = llog_changelog_cancel;
+ rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT,
+ obd, &changelog_orig_logops);
+ if (rc) {
+ CERROR("%s: changelog llog setup failed: rc = %d\n",
+ obd->obd_name, rc);
+ RETURN(rc);
+ }
+
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+ LASSERT(ctxt);
+
+ rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL,
+ CHANGELOG_CATALOG);
+ if (rc)
+ GOTO(out_cleanup, rc);
+
+ ctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
+ ctxt->loc_handle->lgh_logops->lop_declare_add =
+ llog_cat_declare_add_rec;
+
+ rc = llog_cat_init_and_process(env, ctxt->loc_handle);
+ if (rc)
+ GOTO(out_close, rc);
+
+ rc = llog_cat_reverse_process(env, ctxt->loc_handle,
+ changelog_init_cb, mdd);
+
+ if (rc < 0) {
+ CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name,
+ rc);
+ GOTO(out_close, rc);
+ }
+
+ CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n",
+ mdd->mdd_cl.mc_index);
+
+ /* setup user changelog */
+ rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT,
+ obd, &changelog_orig_logops);
+ if (rc) {
+ CERROR("%s: changelog users llog setup failed: rc = %d\n",
+ obd->obd_name, rc);
+ GOTO(out_close, rc);
+ }
+
+ uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
+ LASSERT(ctxt);
+
+ rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL,
+ CHANGELOG_USERS);
+ if (rc)
+ GOTO(out_ucleanup, rc);
+
+ uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec;
+ uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec;
+
+ rc = llog_cat_init_and_process(env, uctxt->loc_handle);
+ if (rc)
+ GOTO(out_uclose, rc);
+
+ rc = llog_cat_reverse_process(env, uctxt->loc_handle,
+ changelog_user_init_cb, mdd);
+ if (rc < 0) {
+ CERROR("%s: changelog user init failed: rc = %d\n",
+ obd->obd_name, rc);
+ GOTO(out_uclose, rc);
+ }
+
+ /* If we have registered users, assume we want changelogs on */
+ if (mdd->mdd_cl.mc_lastuser > 0) {
+ rc = mdd_changelog_on(env, mdd, 1);
+ if (rc < 0)
+ GOTO(out_uclose, rc);
+ }
+ llog_ctxt_put(ctxt);
+ llog_ctxt_put(uctxt);
+ RETURN(0);
+out_uclose:
+ llog_cat_close(env, uctxt->loc_handle);
+out_ucleanup:
+ llog_cleanup(env, uctxt);
+out_close:
+ llog_cat_close(env, ctxt->loc_handle);
+out_cleanup:
+ llog_cleanup(env, ctxt);
+ return rc;
+}
+
+static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd)
+{
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ int rc;
+
+ mdd->mdd_cl.mc_index = 0;
+ spin_lock_init(&mdd->mdd_cl.mc_lock);
+ mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+ mdd->mdd_cl.mc_flags = 0; /* off by default */
+ mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK;
+ spin_lock_init(&mdd->mdd_cl.mc_user_lock);
+ mdd->mdd_cl.mc_lastuser = 0;
+
+ rc = mdd_changelog_llog_init(env, mdd);
+ if (rc) {
+ CERROR("%s: changelog setup during init failed: rc = %d\n",
+ obd->obd_name, rc);
+ mdd->mdd_cl.mc_flags |= CLM_ERR;
+ }
+
+ return rc;
+}
+
+static void mdd_changelog_fini(const struct lu_env *env,
+ struct mdd_device *mdd)
+{
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_ctxt *ctxt;
+
+ mdd->mdd_cl.mc_flags = 0;
+
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+ if (ctxt) {
+ llog_cat_close(env, ctxt->loc_handle);
+ llog_cleanup(env, ctxt);
+ }
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT);
+ if (ctxt) {
+ llog_cat_close(env, ctxt->loc_handle);
+ llog_cleanup(env, ctxt);
+ }
+}
+
+int mdd_changelog_write_header(const struct lu_env *env,
+ struct mdd_device *mdd, int markerflags);
+
+/* Start / stop recording */
+int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on)
+{
+ int rc = 0;
+
+ if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) {
+ LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name);
+ if (mdd->mdd_cl.mc_flags & CLM_ERR) {
+ CERROR("Changelogs cannot be enabled due to error "
+ "condition (see %s log).\n",
+ mdd2obd_dev(mdd)->obd_name);
+ rc = -ESRCH;
+ } else {
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ mdd->mdd_cl.mc_flags |= CLM_ON;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+ rc = mdd_changelog_write_header(env, mdd, CLM_START);
+ }
+ } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) {
+ LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name);
+ rc = mdd_changelog_write_header(env, mdd, CLM_FINI);
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ mdd->mdd_cl.mc_flags &= ~CLM_ON;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+ }
+ return rc;
+}
+
+/** Remove entries with indicies up to and including \a endrec from the
+ * changelog
+ * \param mdd
+ * \param endrec
+ * \retval 0 ok
+ */
+int mdd_changelog_llog_cancel(const struct lu_env *env,
+ struct mdd_device *mdd, long long endrec)
+{
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_ctxt *ctxt;
+ long long unsigned cur;
+ int rc;
+
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+ if (ctxt == NULL)
+ return -ENXIO;
+
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ cur = (long long)mdd->mdd_cl.mc_index;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+ if (endrec > cur)
+ endrec = cur;
+
+ /* purge to "0" is shorthand for everything */
+ if (endrec == 0)
+ endrec = cur;
+
+ /* If purging all records, write a header entry so we don't have an
+ empty catalog and we're sure to have a valid starting index next
+ time. In case of crash, we just restart with old log so we're
+ allright. */
+ if (endrec == cur) {
+ /* XXX: transaction is started by llog itself */
+ rc = mdd_changelog_write_header(env, mdd, CLM_PURGE);
+ if (rc)
+ goto out;
+ }
+
+ /* Some records were purged, so reset repeat-access time (so we
+ record new mtime update records, so users can see a file has been
+ changed since the last purge) */
+ mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+
+ rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0);
+out:
+ llog_ctxt_put(ctxt);
+ return rc;
+}
+
+/** Add a CL_MARK record to the changelog
+ * \param mdd
+ * \param markerflags - CLM_*
+ * \retval 0 ok
+ */
+int mdd_changelog_write_header(const struct lu_env *env,
+ struct mdd_device *mdd, int markerflags)
+{
+ struct obd_device *obd = mdd2obd_dev(mdd);
+ struct llog_changelog_rec *rec;
+ struct lu_buf *buf;
+ struct llog_ctxt *ctxt;
+ int reclen;
+ int len = strlen(obd->obd_name);
+ int rc;
+
+ ENTRY;
+
+ if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) {
+ mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+ RETURN(0);
+ }
+
+ reclen = llog_data_len(sizeof(*rec) + len);
+ buf = mdd_buf_alloc(env, reclen);
+ if (buf->lb_buf == NULL)
+ RETURN(-ENOMEM);
+ rec = buf->lb_buf;
+
+ rec->cr.cr_flags = CLF_VERSION;
+ rec->cr.cr_type = CL_MARK;
+ rec->cr.cr_namelen = len;
+ memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen);
+ /* Status and action flags */
+ rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags;
+ rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen);
+ rec->cr_hdr.lrh_type = CHANGELOG_REC;
+ rec->cr.cr_time = cl_time();
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ rec->cr.cr_index = ++mdd->mdd_cl.mc_index;
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+
+ ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
+ LASSERT(ctxt);
+
+ rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL);
+ if (rc > 0)
+ rc = 0;
+ llog_ctxt_put(ctxt);
+
+ /* assume on or off event; reset repeat-access time */
+ mdd->mdd_cl.mc_starttime = cfs_time_current_64();
+ RETURN(rc);
+}
+
+/**
+ * Create ".lustre" directory.
+ */
+static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m)
+{
+ struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
+ struct md_object *mdo;
+ int rc;
+
+ memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid));
+ mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child,
+ mdd_root_dir_name, dot_lustre_name,
+ fid, &dt_directory_features);
+ /* .lustre dir may be already present */
+ if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) {
+ rc = PTR_ERR(mdo);
+ CERROR("creating obj [%s] fid = "DFID" rc = %d\n",
+ dot_lustre_name, PFID(fid), rc);
+ return rc;
+ }
+
+ if (!IS_ERR(mdo))
+ lu_object_put(env, &mdo->mo_lu);
+
+ return 0;
+}
+
+
+static int dot_lustre_mdd_permission(const struct lu_env *env,
+ struct md_object *pobj,
+ struct md_object *cobj,
+ struct md_attr *attr, int mask)
+{
+ if (mask & ~(MAY_READ | MAY_EXEC))
+ return -EPERM;
+ else
+ return 0;
+}
+
+static int dot_lustre_mdd_xattr_get(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf,
+ const char *name)
+{
+ return 0;
+}
+
+static int dot_lustre_mdd_xattr_list(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf)
+{
+ return 0;
+}
+
+static int dot_lustre_mdd_xattr_set(const struct lu_env *env,
+ struct md_object *obj,
+ const struct lu_buf *buf, const char *name,
+ int fl)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_xattr_del(const struct lu_env *env,
+ struct md_object *obj,
+ const char *name)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_readlink(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf)
+{
+ return 0;
+}
+
+static int dot_lustre_mdd_object_create(const struct lu_env *env,
+ struct md_object *obj,
+ const struct md_op_spec *spec,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_ref_add(const struct lu_env *env,
+ struct md_object *obj,
+ const struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_ref_del(const struct lu_env *env,
+ struct md_object *obj,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj,
+ int flags)
+{
+ struct mdd_object *mdd_obj = md2mdd_obj(obj);
+
+ mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
+ mdd_obj->mod_count++;
+ mdd_write_unlock(env, mdd_obj);
+
+ return 0;
+}
+
+static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj,
+ struct md_attr *ma, int mode)
+{
+ struct mdd_object *mdd_obj = md2mdd_obj(obj);
+
+ mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
+ mdd_obj->mod_count--;
+ mdd_write_unlock(env, mdd_obj);
+
+ return 0;
+}
+
+static int dot_lustre_mdd_object_sync(const struct lu_env *env,
+ struct md_object *obj)
+{
+ return -ENOSYS;
+}
+
+static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj,
+ char *path, int pathlen, __u64 *recno, int *linkno)
+{
+ return -ENOSYS;
+}
+
+static int dot_file_lock(const struct lu_env *env, struct md_object *obj,
+ struct lov_mds_md *lmm, struct ldlm_extent *extent,
+ struct lustre_handle *lockh)
+{
+ return -ENOSYS;
+}
+
+static int dot_file_unlock(const struct lu_env *env, struct md_object *obj,
+ struct lov_mds_md *lmm, struct lustre_handle *lockh)
+{
+ return -ENOSYS;
+}
+
+static struct md_object_operations mdd_dot_lustre_obj_ops = {
+ .moo_permission = dot_lustre_mdd_permission,
+ .moo_attr_get = mdd_attr_get,
+ .moo_attr_set = mdd_attr_set,
+ .moo_xattr_get = dot_lustre_mdd_xattr_get,
+ .moo_xattr_list = dot_lustre_mdd_xattr_list,
+ .moo_xattr_set = dot_lustre_mdd_xattr_set,
+ .moo_xattr_del = dot_lustre_mdd_xattr_del,
+ .moo_readpage = mdd_readpage,
+ .moo_readlink = dot_lustre_mdd_readlink,
+ .moo_object_create = dot_lustre_mdd_object_create,
+ .moo_ref_add = dot_lustre_mdd_ref_add,
+ .moo_ref_del = dot_lustre_mdd_ref_del,
+ .moo_open = dot_lustre_mdd_open,
+ .moo_close = dot_lustre_mdd_close,
+ .moo_capa_get = mdd_capa_get,
+ .moo_object_sync = dot_lustre_mdd_object_sync,
+ .moo_path = dot_lustre_mdd_path,
+ .moo_file_lock = dot_file_lock,
+ .moo_file_unlock = dot_file_unlock,
+};
+
+
+static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p,
+ const struct lu_name *lname, struct lu_fid *f,
+ struct md_op_spec *spec)
+{
+ if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0)
+ *f = LU_OBF_FID;
+ else
+ return -ENOENT;
+
+ return 0;
+}
+
+static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env,
+ struct md_object *obj,
+ mdl_mode_t mode)
+{
+ return MDL_MINMODE;
+}
+
+static int dot_lustre_mdd_create(const struct lu_env *env,
+ struct md_object *pobj,
+ const struct lu_name *lname,
+ struct md_object *child,
+ struct md_op_spec *spec,
+ struct md_attr* ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_create_data(const struct lu_env *env,
+ struct md_object *p,
+ struct md_object *o,
+ const struct md_op_spec *spec,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_rename(const struct lu_env *env,
+ struct md_object *src_pobj,
+ struct md_object *tgt_pobj,
+ const struct lu_fid *lf,
+ const struct lu_name *lsname,
+ struct md_object *tobj,
+ const struct lu_name *ltname,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_link(const struct lu_env *env,
+ struct md_object *tgt_obj,
+ struct md_object *src_obj,
+ const struct lu_name *lname,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_unlink(const struct lu_env *env,
+ struct md_object *pobj,
+ struct md_object *cobj,
+ const struct lu_name *lname,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_name_insert(const struct lu_env *env,
+ struct md_object *obj,
+ const struct lu_name *lname,
+ const struct lu_fid *fid,
+ const struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_name_remove(const struct lu_env *env,
+ struct md_object *obj,
+ const struct lu_name *lname,
+ const struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int dot_lustre_mdd_rename_tgt(const struct lu_env *env,
+ struct md_object *pobj,
+ struct md_object *tobj,
+ const struct lu_fid *fid,
+ const struct lu_name *lname,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+
+static struct md_dir_operations mdd_dot_lustre_dir_ops = {
+ .mdo_is_subdir = mdd_is_subdir,
+ .mdo_lookup = dot_lustre_mdd_lookup,
+ .mdo_lock_mode = dot_lustre_mdd_lock_mode,
+ .mdo_create = dot_lustre_mdd_create,
+ .mdo_create_data = dot_lustre_mdd_create_data,
+ .mdo_rename = dot_lustre_mdd_rename,
+ .mdo_link = dot_lustre_mdd_link,
+ .mdo_unlink = dot_lustre_mdd_unlink,
+ .mdo_name_insert = dot_lustre_mdd_name_insert,
+ .mdo_name_remove = dot_lustre_mdd_name_remove,
+ .mdo_rename_tgt = dot_lustre_mdd_rename_tgt,
+};
+
+static int obf_attr_get(const struct lu_env *env, struct md_object *obj,
+ struct md_attr *ma)
+{
+ struct mdd_device *mdd = mdo2mdd(obj);
+
+ /* "fid" is a virtual object and hence does not have any "real"
+ * attributes. So we reuse attributes of .lustre for "fid" dir */
+ return mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma);
+}
+
+static int obf_attr_set(const struct lu_env *env, struct md_object *obj,
+ const struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int obf_xattr_list(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf)
+{
+ return 0;
+}
+
+static int obf_xattr_get(const struct lu_env *env,
+ struct md_object *obj, struct lu_buf *buf,
+ const char *name)
+{
+ struct mdd_device *mdd = mdo2mdd(obj);
+ struct mdd_object *root;
+ struct lu_fid rootfid;
+ int rc = 0;
+
+ /*
+ * .lustre returns default striping which is 'stored'
+ * in the root
+ */
+ if (strcmp(name, XATTR_NAME_LOV) == 0) {
+ rc = dt_root_get(env, mdd->mdd_child, &rootfid);
+ if (rc)
+ return rc;
+ root = mdd_object_find(env, mdd, &rootfid);
+ if (IS_ERR(root))
+ return PTR_ERR(root);
+ rc = mdo_xattr_get(env, root, buf, name,
+ mdd_object_capa(env, md2mdd_obj(obj)));
+ mdd_object_put(env, root);
+ }
+
+ return rc;
+}
+
+static int obf_xattr_set(const struct lu_env *env,
+ struct md_object *obj,
+ const struct lu_buf *buf, const char *name,
+ int fl)
+{
+ return -EPERM;
+}
+
+static int obf_xattr_del(const struct lu_env *env,
+ struct md_object *obj,
+ const char *name)
+{
+ return -EPERM;
+}
+
+static int obf_mdd_open(const struct lu_env *env, struct md_object *obj,
+ int flags)
+{
+ struct mdd_object *mdd_obj = md2mdd_obj(obj);
+
+ mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
+ mdd_obj->mod_count++;
+ mdd_write_unlock(env, mdd_obj);
+
+ return 0;
+}
+
+static int obf_mdd_close(const struct lu_env *env, struct md_object *obj,
+ struct md_attr *ma, int mode)
+{
+ struct mdd_object *mdd_obj = md2mdd_obj(obj);
+
+ mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
+ mdd_obj->mod_count--;
+ mdd_write_unlock(env, mdd_obj);
+
+ return 0;
+}
+
+/** Nothing to list in "fid" directory */
+static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj,
+ const struct lu_rdpg *rdpg)
+{
+ return -EPERM;
+}
+
+static int obf_path(const struct lu_env *env, struct md_object *obj,
+ char *path, int pathlen, __u64 *recno, int *linkno)
+{
+ return -ENOSYS;
+}
+
+static struct md_object_operations mdd_obf_obj_ops = {
+ .moo_attr_get = obf_attr_get,
+ .moo_attr_set = obf_attr_set,
+ .moo_xattr_list = obf_xattr_list,
+ .moo_xattr_get = obf_xattr_get,
+ .moo_xattr_set = obf_xattr_set,
+ .moo_xattr_del = obf_xattr_del,
+ .moo_open = obf_mdd_open,
+ .moo_close = obf_mdd_close,
+ .moo_readpage = obf_mdd_readpage,
+ .moo_path = obf_path
+};
+
+/**
+ * Lookup method for "fid" object. Only filenames with correct SEQ:OID format
+ * are valid. We also check if object with passed fid exists or not.
+ */
+static int obf_lookup(const struct lu_env *env, struct md_object *p,
+ const struct lu_name *lname, struct lu_fid *f,
+ struct md_op_spec *spec)
+{
+ char *name = (char *)lname->ln_name;
+ struct mdd_device *mdd = mdo2mdd(p);
+ struct mdd_object *child;
+ int rc = 0;
+
+ while (*name == '[')
+ name++;
+
+ sscanf(name, SFID, RFID(f));
+ if (!fid_is_sane(f)) {
+ CWARN("%s: bad FID format [%s], should be "DFID"\n",
+ mdd2obd_dev(mdd)->obd_name, lname->ln_name,
+ (__u64)FID_SEQ_NORMAL, 1, 0);
+ GOTO(out, rc = -EINVAL);
+ }
+
+ if (!fid_is_norm(f)) {
+ CWARN("%s: "DFID" is invalid, sequence should be "
+ ">= "LPX64"\n", mdd2obd_dev(mdd)->obd_name, PFID(f),
+ (__u64)FID_SEQ_NORMAL);
+ GOTO(out, rc = -EINVAL);
+ }
+
+ /* Check if object with this fid exists */
+ child = mdd_object_find(env, mdd, f);
+ if (child == NULL)
+ GOTO(out, rc = 0);
+ if (IS_ERR(child))
+ GOTO(out, rc = PTR_ERR(child));
+
+ if (mdd_object_exists(child) == 0)
+ rc = -ENOENT;
+
+ mdd_object_put(env, child);
+
+out:
+ return rc;
+}
+
+static int obf_create(const struct lu_env *env, struct md_object *pobj,
+ const struct lu_name *lname, struct md_object *child,
+ struct md_op_spec *spec, struct md_attr* ma)
+{
+ return -EPERM;
+}
+
+static int obf_rename(const struct lu_env *env,
+ struct md_object *src_pobj, struct md_object *tgt_pobj,
+ const struct lu_fid *lf, const struct lu_name *lsname,
+ struct md_object *tobj, const struct lu_name *ltname,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int obf_link(const struct lu_env *env, struct md_object *tgt_obj,
+ struct md_object *src_obj, const struct lu_name *lname,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static int obf_unlink(const struct lu_env *env, struct md_object *pobj,
+ struct md_object *cobj, const struct lu_name *lname,
+ struct md_attr *ma)
+{
+ return -EPERM;
+}
+
+static struct md_dir_operations mdd_obf_dir_ops = {
+ .mdo_lookup = obf_lookup,
+ .mdo_create = obf_create,
+ .mdo_rename = obf_rename,
+ .mdo_link = obf_link,
+ .mdo_unlink = obf_unlink
+};
+
+/**
+ * Create special in-memory "fid" object for open-by-fid.
+ */
+static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m)
+{
+ struct mdd_object *mdd_obf;
+ struct lu_object *obf_lu_obj;
+ int rc = 0;
+
+ m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m,
+ &LU_OBF_FID);
+ if (m->mdd_dot_lustre_objs.mdd_obf == NULL ||
+ IS_ERR(m->mdd_dot_lustre_objs.mdd_obf))
+ GOTO(out, rc = -ENOENT);
+
+ mdd_obf = m->mdd_dot_lustre_objs.mdd_obf;
+ mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops;
+ mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops;
+ /* Don't allow objects to be created in "fid" dir */
+ mdd_obf->mod_flags |= IMMUTE_OBJ;
+
+ obf_lu_obj = mdd2lu_obj(mdd_obf);
+ obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR);
+
+out:
+ return rc;
+}
+
+/** Setup ".lustre" directory object */
+static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m)
+{
+ struct dt_object *dt_dot_lustre;
+ struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
+ int rc;
+
+ rc = create_dot_lustre_dir(env, m);
+ if (rc)
+ return rc;
+
+ dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name,
+ dot_lustre_name, fid);
+ if (IS_ERR(dt_dot_lustre)) {
+ rc = PTR_ERR(dt_dot_lustre);
+ GOTO(out, rc);
+ }
+
+ /* references are released in mdd_device_shutdown() */
+ m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header,
+ &mdd_device_type));
+
+ m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops;
+ m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops;
+
+ rc = mdd_obf_setup(env, m);
+ if (rc)
+ CERROR("Error initializing \"fid\" object - %d.\n", rc);
+
+out:
+ RETURN(rc);
+}
+