- OBD_ALLOC_PTR(ori);
- if (ori == NULL)
- RETURN(-ENOMEM);
-
- mutex_lock(&osp_register_list_lock);
-
- osp = class_name2obd(ospname);
- if (osp != NULL && osp->obd_set_up == 1) {
- struct obd_uuid *uuid;
-
- OBD_ALLOC_PTR(uuid);
- if (uuid == NULL) {
- mutex_unlock(&osp_register_list_lock);
- OBD_FREE_PTR(ori);
- RETURN(-ENOMEM);
- }
- memcpy(uuid->uuid, ospname, strlen(ospname));
- *exp = cfs_hash_lookup(osp->obd_uuid_hash, uuid);
- OBD_FREE_PTR(uuid);
- }
-
- memcpy(ori->ori_name, ospname, strlen(ospname));
- ori->ori_exp = exp;
- ori->ori_cb_func = cb_func;
- ori->ori_cb_data = cb_data;
- CFS_INIT_LIST_HEAD(&ori->ori_list);
- cfs_list_add(&ori->ori_list, &osp_register_list);
-
- if (*exp != NULL && cb_func != NULL)
- cb_func(cb_data);
-
- mutex_unlock(&osp_register_list_lock);
- RETURN(0);
-}
-EXPORT_SYMBOL(lustre_register_osp_item);
-
-void lustre_deregister_osp_item(struct obd_export **exp)
-{
- struct osp_register_item *ori, *tmp;
-
- mutex_lock(&osp_register_list_lock);
- cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) {
- if (exp == ori->ori_exp) {
- if (*exp)
- class_export_put(*exp);
- cfs_list_del(&ori->ori_list);
- OBD_FREE_PTR(ori);
- break;
- }
- }
- mutex_unlock(&osp_register_list_lock);
-}
-EXPORT_SYMBOL(lustre_deregister_osp_item);
-
-static void lustre_notify_osp_list(struct obd_export *exp)
-{
- struct osp_register_item *ori, *tmp;
- LASSERT(exp != NULL);
-
- mutex_lock(&osp_register_list_lock);
- cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) {
- if (strcmp(exp->exp_obd->obd_name, ori->ori_name))
- continue;
- if (*ori->ori_exp != NULL)
- continue;
- *ori->ori_exp = class_export_get(exp);
- if (ori->ori_cb_func != NULL)
- ori->ori_cb_func(ori->ori_cb_data);
- }
- mutex_unlock(&osp_register_list_lock);
-}
-
-static int lustre_osp_connect(struct obd_device *osp)
-{
- struct lu_env env;
- struct lu_context session_ctx;
- struct obd_export *exp;
- struct obd_uuid *uuid = NULL;
- struct obd_connect_data *data = NULL;
- int rc;
- ENTRY;
-
- /* log has been fully processed, let clients connect */
- rc = lu_env_init(&env, osp->obd_lu_dev->ld_type->ldt_ctx_tags);
- if (rc != 0)
- RETURN(rc);
-
- lu_context_init(&session_ctx, LCT_SESSION);
- session_ctx.lc_thread = NULL;
- lu_context_enter(&session_ctx);
- env.le_ses = &session_ctx;
-
- OBD_ALLOC_PTR(data);
- if (data == NULL)
- GOTO(out, rc = -ENOMEM);
-
- data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
- data->ocd_version = LUSTRE_VERSION_CODE;
- data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
- OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
- OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
- OBD_CONNECT_LIGHTWEIGHT;
- OBD_ALLOC_PTR(uuid);
- if (uuid == NULL)
- GOTO(out, rc = -ENOMEM);
-
- if (strlen(osp->obd_name) > sizeof(uuid->uuid)) {
- CERROR("%s: Too long osp name %s, max_size is %d\n",
- osp->obd_name, osp->obd_name, (int)sizeof(uuid->uuid));
- GOTO(out, rc = -EINVAL);
- }
- /* Use osp name as the uuid, so we find the export by
- * osp name later */
- memcpy(uuid->uuid, osp->obd_name, strlen(osp->obd_name));
- rc = obd_connect(&env, &exp, osp, uuid, data, NULL);
- if (rc != 0)
- CERROR("%s: connect failed: rc = %d\n", osp->obd_name, rc);
- else
- lustre_notify_osp_list(exp);
-
-out:
- if (data != NULL)
- OBD_FREE_PTR(data);
- if (uuid != NULL)
- OBD_FREE_PTR(uuid);
-
- lu_env_fini(&env);
- lu_context_exit(&session_ctx);
- lu_context_fini(&session_ctx);
-
- RETURN(rc);
-}
-
-/**
- * osp-on-ost is used by slaves (Non-MDT0 targets) to manage the connection
- * to MDT0.
- *
- * The OSTs will communicate with MDT0 by the connection established by the
- * osp-on-ost to get quota and fid sequence.
- *
- **/
-static int lustre_osp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
-{
- struct obd_connect_data *data = NULL;
- struct obd_device *obd;
- char *ospname = NULL;
- char *ospuuid = NULL;
- int rc;
- ENTRY;
-
- rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
- lcfg->lcfg_nid);
- if (rc) {
- CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
- GOTO(out, rc);
- }
-
- OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
- if (ospname == NULL)
- GOTO(out, rc = -ENOMEM);
-
- rc = tgt_name2ospname(lsi->lsi_svname, ospname);
- if (rc != 0) {
- CERROR("%s change ospname error: rc %d\n",
- lsi->lsi_svname, rc);
- GOTO(out, rc);
- }
-
- OBD_ALLOC(ospuuid, MTI_NAME_MAXLEN);
- if (ospuuid == NULL)
- GOTO(out, rc = -ENOMEM);
-
- sprintf(ospuuid, "%s_UUID", ospname);
- rc = lustre_start_simple(ospname, LUSTRE_OSP_NAME,
- ospuuid, lustre_cfg_string(lcfg, 1),
- 0, 0, 0);
- if (rc) {
- CERROR("%s: setup up failed: rc %d\n", ospname, rc);
- GOTO(out, rc);
- }
-
- obd = class_name2obd(ospname);
- LASSERT(obd != NULL);
-
- rc = lustre_osp_connect(obd);
- if (rc != 0)
- CERROR("%s: connect failed: rc = %d\n", ospname, rc);
-out:
- if (data != NULL)
- OBD_FREE_PTR(data);
- if (ospname != NULL)
- OBD_FREE(ospname, MTI_NAME_MAXLEN);
- if (ospuuid != NULL)
- OBD_FREE(ospuuid, MTI_NAME_MAXLEN);
-
- RETURN(rc);
-}
-
-static int lustre_osp_add_conn(struct lustre_cfg *cfg,
- struct lustre_sb_info *lsi)
-{
- struct lustre_cfg_bufs *bufs = NULL;
- struct lustre_cfg *lcfg = NULL;
- char *ospname = NULL;
- struct obd_device *osp;
- int rc;
- ENTRY;
-
- OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
- if (ospname == NULL)
- GOTO(out, rc = -ENOMEM);
-
- rc = tgt_name2ospname(lsi->lsi_svname, ospname);
- if (rc != 0) {
- CERROR("%s change ospname error: rc %d\n",
- lsi->lsi_svname, rc);
- GOTO(out, rc);
- }
-
- OBD_ALLOC_PTR(bufs);
- if (bufs == NULL)
- GOTO(out, rc = -ENOMEM);
-
- lustre_cfg_bufs_reset(bufs, ospname);
- lustre_cfg_bufs_set_string(bufs, 1,
- lustre_cfg_string(cfg, 1));
-
- lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
-
- osp = class_name2obd(ospname);
- if (osp == NULL) {
- CERROR("Can not find %s\n", ospname);
- GOTO(out, rc = -EINVAL);
- }
-
- rc = class_add_conn(osp, lcfg);
- if (rc)
- CERROR("%s: can't add conn: rc = %d\n", ospname, rc);
-
-out:
- if (bufs != NULL)
- OBD_FREE_PTR(bufs);
- if (lcfg != NULL)
- lustre_cfg_free(lcfg);
- if (ospname != NULL)
- OBD_FREE(ospname, MTI_NAME_MAXLEN);
-
- RETURN(rc);
-}
-
-/**
- * Retrieve MDT nids from the client log, then start the osp-on-ost device.
- * there are only two scenarios which would include mdt nid.
- * 1.
- * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
- * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
- * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
- * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
- * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
- * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
- * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
- * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
- * 2.
- * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
- * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
- * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
- * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
-**/
-static int client_osp_config_process(const struct lu_env *env,
- struct llog_handle *handle,
- struct llog_rec_hdr *rec, void *data)
-{
- struct config_llog_instance *clli = data;
- int cfg_len = rec->lrh_len;
- char *cfg_buf = (char *) (rec + 1);
- struct lustre_cfg *lcfg = NULL;
- struct lustre_sb_info *lsi;
- int rc = 0, swab = 0;
- ENTRY;
-
- if (rec->lrh_type != OBD_CFG_REC) {
- CERROR("Unknown llog record type %#x encountered\n",
- rec->lrh_type);
- RETURN(-EINVAL);
- }
-
- LASSERT(clli->cfg_sb != NULL);
- lsi = s2lsi(clli->cfg_sb);
-
- lcfg = (struct lustre_cfg *)cfg_buf;
- if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
- lustre_swab_lustre_cfg(lcfg);
- swab = 1;
- }
-
- rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
- if (rc)
- GOTO(out, rc);
-
- switch (lcfg->lcfg_command) {
- case LCFG_MARKER: {
- struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
-
- lustre_swab_cfg_marker(marker, swab,
- LUSTRE_CFG_BUFLEN(lcfg, 1));
- if (marker->cm_flags & CM_SKIP ||
- marker->cm_flags & CM_EXCLUDE)
- GOTO(out, rc = 0);
-
- if (!tgt_is_mdt0(marker->cm_tgtname))
- GOTO(out, rc = 0);
-
- if(!strncmp(marker->cm_comment, "add mdc", 7) ||
- !strncmp(marker->cm_comment, "add failnid", 11)) {
- if (marker->cm_flags & CM_START) {
- clli->cfg_flags = CFG_F_MARKER;
- /* This hack is to differentiate the
- * ADD_UUID is come from "add mdc" record
- * or from "add failnid" record. */
- if (!strncmp(marker->cm_comment,
- "add failnid", 11))
- clli->cfg_flags |= CFG_F_SKIP;
- } else if (marker->cm_flags & CM_END) {
- clli->cfg_flags = 0;
- }
- }
- break;
- }
- case LCFG_ADD_UUID: {
- if (clli->cfg_flags == CFG_F_MARKER) {
- rc = lustre_osp_setup(lcfg, lsi);
- /* XXX: process only the first nid as
- * we don't need another instance of osp */
- clli->cfg_flags |= CFG_F_SKIP;
- } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
- rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
- lcfg->lcfg_nid);
- if (rc)
- CERROR("%s: Fail to add uuid, rc:%d\n",
- lsi->lsi_svname, rc);
- }
- break;
- }
- case LCFG_ADD_CONN: {
- if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
- rc = lustre_osp_add_conn(lcfg, lsi);
- break;
- }
- default:
- break;
- }
-out:
- RETURN(rc);
-}
-
-static int lustre_disconnect_osp(struct super_block *sb)
-{
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct obd_device *osp;
- char *ospname = NULL;
- char *logname = NULL;
- struct lustre_cfg *lcfg = NULL;
- struct lustre_cfg_bufs *bufs = NULL;
- struct config_llog_instance *cfg = NULL;
- int rc;
- ENTRY;
-
- LASSERT(IS_OST(lsi) || IS_MDT(lsi));
- if (IS_MDT(lsi)) {
- int index;
-
- /* Only disconnect MDT0-osp-MDT0 here, other osp on MDT
- * will be disconnect during MDT stack cleanup.
- * FIXME: remove later when quota on DNE is finished */
- rc = server_name2index(lsi->lsi_svname, &index, NULL);
- if (rc < 0)
- RETURN(rc);
- if (index != 0)
- RETURN(0);
- }
- OBD_ALLOC(logname, MTI_NAME_MAXLEN);
- if (logname == NULL)
- RETURN(-ENOMEM);
-
- OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
- if (ospname == NULL)
- GOTO(out, rc = -ENOMEM);
-
- rc = server_name2fsname(lsi->lsi_svname, ospname, NULL);
- if (rc != 0) {
- CERROR("%s: get fsname error: %d\n",
- lsi->lsi_svname, rc);
- GOTO(out, rc);
- }
- sprintf(logname, "%s-client", ospname);
-
- OBD_ALLOC_PTR(cfg);
- if (cfg == NULL)
- GOTO(out, rc = -ENOMEM);
-
- /* end log first */
- cfg->cfg_instance = sb;
- rc = lustre_end_log(sb, logname, cfg);
- if (rc != 0) {
- CERROR("Can't end config log %s\n", ospname);
- GOTO(out, rc);
- }
-
- rc = tgt_name2ospname(lsi->lsi_svname, ospname);
- if (rc != 0) {
- CERROR("%s: get osp name error: %d\n",
- lsi->lsi_svname, rc);
- GOTO(out, rc);
- }
-
- osp = class_name2obd(ospname);
- if (osp == NULL) {
- CERROR("Can't find osp-on-ost %s\n", ospname);
- GOTO(out, rc = -ENOENT);
- }
-
- OBD_ALLOC_PTR(bufs);
- if (bufs == NULL)
- GOTO(out, rc = -ENOMEM);
-
- lustre_cfg_bufs_reset(bufs, osp->obd_name);
- lustre_cfg_bufs_set_string(bufs, 1, NULL);
- lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
- if (!lcfg)
- GOTO(out, rc = -ENOMEM);
-
- /* Disconnect import first. NULL is passed for the '@env', since
- * it will not be used for the 'osp-on-ost'. (see osp_shutdown()) */
- rc = osp->obd_lu_dev->ld_ops->ldo_process_config(NULL, osp->obd_lu_dev,
- lcfg);
-out:
- if (lcfg)
- lustre_cfg_free(lcfg);
- if (bufs)
- OBD_FREE_PTR(bufs);
- if (cfg)
- OBD_FREE_PTR(cfg);
- if (ospname)
- OBD_FREE(ospname, MTI_NAME_MAXLEN);
- if (logname)
- OBD_FREE(logname, MTI_NAME_MAXLEN);
- RETURN(rc);
-}
-
-/**
- * Stop the osp(fsname-MDT0000-osp-OSTxxxx) or (fsname-MDT0000-osp-MDT0000) for an OST target.
- **/
-static int lustre_stop_osp(struct super_block *sb)
-{
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct obd_device *osp = NULL;
- char *ospname = NULL;
- int rc = 0;
- ENTRY;
-
- LASSERT(IS_OST(lsi) || IS_MDT(lsi));
- if (IS_MDT(lsi)) {
- int index;
-
- /* Only disconnect MDT0-osp-MDT0 here, other osp on MDT
- * will be disconnect during MDT stack cleanup.
- * FIXME: remove later when quota on DNE is finished */
- rc = server_name2index(lsi->lsi_svname, &index, NULL);
- if (rc < 0)
- RETURN(rc);
- if (index != 0)
- RETURN(0);
- }
-
- OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
- rc = tgt_name2ospname(lsi->lsi_svname, ospname);
- if (rc != 0) {
- CERROR("%s get fsname error: rc %d\n",
- lsi->lsi_svname, rc);
- GOTO(cleanup, rc);
- }
-
- osp = class_name2obd(ospname);
- if (osp == NULL) {
- CERROR("Can not find osp-on-ost %s\n", ospname);
- GOTO(cleanup, rc = -ENOENT);
- }
-
- osp->obd_force = 1;
- rc = class_manual_cleanup(osp);
-
-cleanup:
- if (ospname != NULL)
- OBD_FREE(ospname, MTI_NAME_MAXLEN);
- RETURN(rc);
-}
-
-DEFINE_MUTEX(server_start_lock);
-
-/* Stop MDS/OSS if nobody is using them */
-static int server_stop_servers(int lsiflags)
-{
- struct obd_device *obd = NULL;
- struct obd_type *type = NULL;
- int rc = 0;
- ENTRY;
-
- mutex_lock(&server_start_lock);
-
- /* Either an MDT or an OST or neither */
- /* if this was an MDT, and there are no more MDT's, clean up the MDS */
- if ((lsiflags & LDD_F_SV_TYPE_MDT) &&
- (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
- type = class_search_type(LUSTRE_MDT_NAME);
- }
- /* if this was an OST, and there are no more OST's, clean up the OSS */
- if ((lsiflags & LDD_F_SV_TYPE_OST) &&
- (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
- type = class_search_type(LUSTRE_OST_NAME);
- }
-
- if (obd && (!type || !type->typ_refcnt)) {
- int err;
- obd->obd_force = 1;
- /* obd_fail doesn't mean much on a server obd */
- err = class_manual_cleanup(obd);
- if (!rc)
- rc = err;
- }
-
- mutex_unlock(&server_start_lock);
-
- RETURN(rc);
-}
-
-int server_mti_print(char *title, struct mgs_target_info *mti)
-{
- PRINT_CMD(PRINT_MASK, "mti %s\n", title);
- PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
- PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
- PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
- PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
- mti->mti_config_ver, mti->mti_flags);
- return(0);
-}
-
-/**
- * Get service name (svname) from string
- * rc < 0 on error
- * if endptr isn't NULL it is set to end of fsname *
- */
-int server_name2svname(char *label, char *svname, char **endptr)
-{
- int rc;
- char *dash;
-
- /* We use server_name2fsname() just for parsing */
- rc = server_name2fsname(label, NULL, &dash);
- if (rc != 0)
- return rc;
-
- if (*dash != '-')
- return -1;
-
- strncpy(svname, dash + 1, MTI_NAME_MAXLEN);
-
- return 0;
-}
-EXPORT_SYMBOL(server_name2svname);
-
-
-/* Get the index from the obd name.
- rc = server type, or
- rc < 0 on error
- if endptr isn't NULL it is set to end of name */
-int server_name2index(char *svname, __u32 *idx, char **endptr)
-{
- unsigned long index;
- int rc;
- char *dash;
-
- /* We use server_name2fsname() just for parsing */
- rc = server_name2fsname(svname, NULL, &dash);
- if (rc != 0)
- return rc;
-
- if (*dash != '-')
- return -EINVAL;
-
- dash++;
-
- if (strncmp(dash, "MDT", 3) == 0)
- rc = LDD_F_SV_TYPE_MDT;
- else if (strncmp(dash, "OST", 3) == 0)
- rc = LDD_F_SV_TYPE_OST;
- else
- return -EINVAL;
-
- dash += 3;
-
- if (strcmp(dash, "all") == 0)
- return rc | LDD_F_SV_ALL;
-
- index = simple_strtoul(dash, endptr, 16);
- *idx = index;
-
- return rc;
-}
-EXPORT_SYMBOL(server_name2index);
-
-/* Generate data for registration */
-static int server_lsi2mti(struct lustre_sb_info *lsi,
- struct mgs_target_info *mti)
-{
- lnet_process_id_t id;
- int rc, i = 0;
- ENTRY;
-
- if (!IS_SERVER(lsi))
- RETURN(-EINVAL);
-
- strncpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname));
-
- mti->mti_nid_count = 0;
- while (LNetGetId(i++, &id) != -ENOENT) {
- if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
- continue;
-
- /* server use --servicenode param, only allow specified
- * nids be registered */
- if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
- class_match_nid(lsi->lsi_lmd->lmd_params,
- PARAM_FAILNODE, id.nid) < 1)
- continue;
-
- /* match specified network */
- if (!class_match_net(lsi->lsi_lmd->lmd_params,
- PARAM_NETWORK, LNET_NIDNET(id.nid)))
- continue;
-
- mti->mti_nids[mti->mti_nid_count] = id.nid;
- mti->mti_nid_count++;
- if (mti->mti_nid_count >= MTI_NIDS_MAX) {
- CWARN("Only using first %d nids for %s\n",
- mti->mti_nid_count, mti->mti_svname);
- break;
- }
- }
-
- mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
- mti->mti_config_ver = 0;
-
- rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
- if (rc != 0)
- return rc;
-
- rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
- if (rc < 0)
- return rc;
- /* Orion requires index to be set */
- LASSERT(!(rc & LDD_F_NEED_INDEX));
- /* keep only LDD flags */
- mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
- if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
- mti->mti_flags |= LDD_F_UPDATE;
- strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
- sizeof(mti->mti_params));
- return 0;
-}
-
-/* Register an old or new target with the MGS. If needed MGS will construct
- startup logs and assign index */
-static int server_register_target(struct lustre_sb_info *lsi)
-{
- struct obd_device *mgc = lsi->lsi_mgc;
- struct mgs_target_info *mti = NULL;
- bool writeconf;
- int rc;
- ENTRY;
-
- LASSERT(mgc);
-
- if (!IS_SERVER(lsi))
- RETURN(-EINVAL);
-
- OBD_ALLOC_PTR(mti);
- if (!mti)
- RETURN(-ENOMEM);
-
- rc = server_lsi2mti(lsi, mti);
- if (rc)
- GOTO(out, rc);
-
- CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
- mti->mti_svname, mti->mti_fsname,
- libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
- mti->mti_flags);
-
- /* if write_conf is true, the registration must succeed */
- writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
- mti->mti_flags |= LDD_F_OPC_REG;
-
- /* Register the target */
- /* FIXME use mgc_process_config instead */
- rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
- sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
- sizeof(*mti), mti, NULL);
- if (rc) {
- if (mti->mti_flags & LDD_F_ERROR) {
- LCONSOLE_ERROR_MSG(0x160,
- "The MGS is refusing to allow this "
- "server (%s) to start. Please see messages"
- " on the MGS node.\n", lsi->lsi_svname);
- } else if (writeconf) {
- LCONSOLE_ERROR_MSG(0x15f,
- "Communication to the MGS return error %d. "
- "Is the MGS running?\n", rc);
- } else {
- CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
- /* reset the error code for non-fatal error. */
- rc = 0;
- }
- GOTO(out, rc);
- }
-
-out:
- if (mti)
- OBD_FREE_PTR(mti);
- RETURN(rc);
-}
-
-/**
- * Notify the MGS that this target is ready.
- * Used by IR - if the MGS receives this message, it will notify clients.
- */
-static int server_notify_target(struct super_block *sb, struct obd_device *obd)
-{
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct obd_device *mgc = lsi->lsi_mgc;
- struct mgs_target_info *mti = NULL;
- int rc;
- ENTRY;
-
- LASSERT(mgc);
-
- if (!(IS_SERVER(lsi)))
- RETURN(-EINVAL);
-
- OBD_ALLOC_PTR(mti);
- if (!mti)
- RETURN(-ENOMEM);
- rc = server_lsi2mti(lsi, mti);
- if (rc)
- GOTO(out, rc);
-
- mti->mti_instance = obd->u.obt.obt_instance;
- mti->mti_flags |= LDD_F_OPC_READY;
-
- /* FIXME use mgc_process_config instead */
- rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
- sizeof(KEY_REGISTER_TARGET),
- KEY_REGISTER_TARGET,
- sizeof(*mti), mti, NULL);
-
- /* Imperative recovery: if the mgs informs us to use IR? */
- if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
- (mti->mti_flags & LDD_F_IR_CAPABLE))
- lsi->lsi_flags |= LDD_F_IR_CAPABLE;
-
-out:
- if (mti)
- OBD_FREE_PTR(mti);
- RETURN(rc);
-
-}
-
-/**
- * Start the osp(fsname-MDT0000-osp-OSTxxxx) for an OST target,
- * which would be used to communicate with MDT0 for quota and FID.
- **/
-static int lustre_start_osp(struct super_block *sb)
-{
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct config_llog_instance *cfg = NULL;
- struct obd_device *osp;
- char *ospname = NULL;
- char *logname = NULL;
- char *tgt;
- int rc;
- ENTRY;
-
- LASSERT(IS_OST(lsi) || IS_MDT(lsi));
- OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
- OBD_ALLOC(logname, MTI_NAME_MAXLEN);
- if (ospname == NULL || logname == NULL)
- GOTO(cleanup, rc = -ENOMEM);
-
- rc = server_name2fsname(lsi->lsi_svname, ospname, &tgt);
- if (rc != 0) {
- CERROR("%s change fsname error: rc %d\n",
- lsi->lsi_svname, rc);
- GOTO(cleanup, rc);
- }
- sprintf(logname, "%s-client", ospname);
-
- rc = tgt_name2ospname(lsi->lsi_svname, ospname);
- if (rc != 0) {
- CERROR("%s change ospname error: rc %d\n",
- lsi->lsi_svname, rc);
- GOTO(cleanup, rc);
- }
-
- osp = class_name2obd(ospname);
- if (osp != NULL)
- GOTO(cleanup, rc = 0);
-
- OBD_ALLOC_PTR(cfg);
- if (cfg == NULL)
- GOTO(cleanup, rc = -ENOMEM);
-
- cfg->cfg_callback = client_osp_config_process;
- cfg->cfg_instance = sb;
-
- rc = lustre_process_log(sb, logname, cfg);
-
-cleanup:
- if (ospname != NULL)
- OBD_FREE(ospname, MTI_NAME_MAXLEN);
- if (logname != NULL)
- OBD_FREE(logname, MTI_NAME_MAXLEN);
- if (cfg != NULL)
- OBD_FREE_PTR(cfg);
-
- RETURN(rc);
-}
-
-/** Start server targets: MDTs and OSTs
- */
-static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
-{
- struct obd_device *obd;
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct config_llog_instance cfg;
- struct lu_env env;
- struct lu_device *dev;
- int rc;
- ENTRY;
-
- CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
-
- if (IS_MDT(lsi)) {
- /* make sure the MDS is started */
- mutex_lock(&server_start_lock);
- obd = class_name2obd(LUSTRE_MDS_OBDNAME);
- if (!obd) {
- rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
- LUSTRE_MDS_NAME,
- LUSTRE_MDS_OBDNAME"_uuid",
- 0, 0, 0, 0);
- if (rc) {
- mutex_unlock(&server_start_lock);
- CERROR("failed to start MDS: %d\n", rc);
- RETURN(rc);
- }
- }
- mutex_unlock(&server_start_lock);
- }
-
- /* If we're an OST, make sure the global OSS is running */
- if (IS_OST(lsi)) {
- /* make sure OSS is started */
- mutex_lock(&server_start_lock);
- obd = class_name2obd(LUSTRE_OSS_OBDNAME);
- if (!obd) {
- rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
- LUSTRE_OSS_NAME,
- LUSTRE_OSS_OBDNAME"_uuid",
- 0, 0, 0, 0);
- if (rc) {
- mutex_unlock(&server_start_lock);
- CERROR("failed to start OSS: %d\n", rc);
- RETURN(rc);
- }
- }
- mutex_unlock(&server_start_lock);
- }
-
- /* Set the mgc fs to our server disk. This allows the MGC to
- * read and write configs locally, in case it can't talk to the MGS. */
- if (lsi->lsi_srv_mnt) {
- rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
- if (rc)
- GOTO(out_stop_service, rc);
- }
-
- /* Register with MGS */
- rc = server_register_target(lsi);
- if (rc)
- GOTO(out_mgc, rc);
-
- /* Let the target look up the mount using the target's name
- (we can't pass the sb or mnt through class_process_config.) */
- rc = server_register_mount(lsi->lsi_svname, sb, mnt);
- if (rc)
- GOTO(out_mgc, rc);
-
- /* Start targets using the llog named for the target */
- memset(&cfg, 0, sizeof(cfg));
- cfg.cfg_callback = class_config_llog_handler;
- rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
- if (rc) {
- CERROR("failed to start server %s: %d\n",
- lsi->lsi_svname, rc);
- /* Do NOT call server_deregister_mount() here. This makes it
- * impossible to find mount later in cleanup time and leaves
- * @lsi and othder stuff leaked. -umka */
- GOTO(out_mgc, rc);
- }
-
- obd = class_name2obd(lsi->lsi_svname);
- if (!obd) {
- CERROR("no server named %s was started\n", lsi->lsi_svname);
- GOTO(out_mgc, rc = -ENXIO);
- }
-
- if (IS_OST(lsi) || IS_MDT(lsi)) {
- rc = lustre_start_osp(sb);
- if (rc) {
- CERROR("%s: failed to start OSP: %d\n",
- lsi->lsi_svname, rc);
- GOTO(out_mgc, rc);
- }
- }
-
- server_notify_target(sb, obd);
-
- /* calculate recovery timeout, do it after lustre_process_log */
- server_calc_timeout(lsi, obd);
-
- /* log has been fully processed */
- obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
-
- /* log has been fully processed, let clients connect */
- dev = obd->obd_lu_dev;
- if (dev && dev->ld_ops->ldo_prepare) {
- rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
- if (rc == 0) {
- struct lu_context session_ctx;
-
- lu_context_init(&session_ctx, LCT_SESSION);
- session_ctx.lc_thread = NULL;
- lu_context_enter(&session_ctx);
- env.le_ses = &session_ctx;
-
- dev->ld_ops->ldo_prepare(&env, NULL, dev);
-
- lu_env_fini(&env);
- lu_context_exit(&session_ctx);
- lu_context_fini(&session_ctx);
- }
- }
-
- /* abort recovery only on the complete stack:
- * many devices can be involved */
- if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
- (OBP(obd, iocontrol))) {
- obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
- NULL, NULL);
- }
-
-out_mgc:
- /* Release the mgc fs for others to use */
- if (lsi->lsi_srv_mnt)
- server_mgc_clear_fs(lsi->lsi_mgc);
-
-out_stop_service:
- if (rc != 0)
- server_stop_servers(lsi->lsi_flags);
-
- RETURN(rc);
-}
-
-/***************** lustre superblock **************/
-
-struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
-{
- struct lustre_sb_info *lsi;
- ENTRY;
-
- OBD_ALLOC_PTR(lsi);
- if (!lsi)
- RETURN(NULL);
- OBD_ALLOC_PTR(lsi->lsi_lmd);
- if (!lsi->lsi_lmd) {
- OBD_FREE_PTR(lsi);
- RETURN(NULL);
- }
-
- lsi->lsi_lmd->lmd_exclude_count = 0;
- lsi->lsi_lmd->lmd_recovery_time_soft = 0;
- lsi->lsi_lmd->lmd_recovery_time_hard = 0;
- s2lsi_nocast(sb) = lsi;
- /* we take 1 extra ref for our setup */
- cfs_atomic_set(&lsi->lsi_mounts, 1);
-
- /* Default umount style */
- lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
-
- RETURN(lsi);
-}
-
-static int lustre_free_lsi(struct super_block *sb)
-{
- struct lustre_sb_info *lsi = s2lsi(sb);
- ENTRY;
-
- LASSERT(lsi != NULL);
- CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
-
- /* someone didn't call server_put_mount. */
- LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
-
- if (lsi->lsi_lmd != NULL) {
- if (lsi->lsi_lmd->lmd_dev != NULL)
- OBD_FREE(lsi->lsi_lmd->lmd_dev,
- strlen(lsi->lsi_lmd->lmd_dev) + 1);
- if (lsi->lsi_lmd->lmd_profile != NULL)
- OBD_FREE(lsi->lsi_lmd->lmd_profile,
- strlen(lsi->lsi_lmd->lmd_profile) + 1);
- if (lsi->lsi_lmd->lmd_mgssec != NULL)
- OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
- strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
- if (lsi->lsi_lmd->lmd_opts != NULL)
- OBD_FREE(lsi->lsi_lmd->lmd_opts,
- strlen(lsi->lsi_lmd->lmd_opts) + 1);
- if (lsi->lsi_lmd->lmd_exclude_count)
- OBD_FREE(lsi->lsi_lmd->lmd_exclude,
- sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
- lsi->lsi_lmd->lmd_exclude_count);
- if (lsi->lsi_lmd->lmd_mgs != NULL)
- OBD_FREE(lsi->lsi_lmd->lmd_mgs,
- strlen(lsi->lsi_lmd->lmd_mgs) + 1);
- if (lsi->lsi_lmd->lmd_osd_type != NULL)
- OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
- strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
- if (lsi->lsi_lmd->lmd_params != NULL)
- OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
-
- OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
- }
-
- LASSERT(lsi->lsi_llsbi == NULL);
- OBD_FREE(lsi, sizeof(*lsi));
- s2lsi_nocast(sb) = NULL;
-
- RETURN(0);
-}
-
-/* The lsi has one reference for every server that is using the disk -
- e.g. MDT, MGS, and potentially MGC */
-static int lustre_put_lsi(struct super_block *sb)
-{
- struct lustre_sb_info *lsi = s2lsi(sb);
- ENTRY;
-
- LASSERT(lsi != NULL);
-
- CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
- if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
- if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
- obd_disconnect(lsi->lsi_osd_exp);
- /* wait till OSD is gone */
- obd_zombie_barrier();
- }
- lustre_free_lsi(sb);
- RETURN(1);
- }
- RETURN(0);
-}
-
-static int lsi_prepare(struct lustre_sb_info *lsi)
-{
- __u32 index;
- int rc;
- ENTRY;
-
- LASSERT(lsi);
- LASSERT(lsi->lsi_lmd);
-
- /* The server name is given as a mount line option */
- if (lsi->lsi_lmd->lmd_profile == NULL) {
- LCONSOLE_ERROR("Can't determine server name\n");
- RETURN(-EINVAL);
- }
-
- if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
- RETURN(-ENAMETOOLONG);
-
- strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
-
- /* Determine osd type */
- if (lsi->lsi_lmd->lmd_osd_type != NULL) {
- if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
- sizeof(lsi->lsi_osd_type))
- RETURN(-ENAMETOOLONG);
-
- strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
- } else {
- strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
- }
-
- /* XXX: a temp. solution for components using fsfilt
- * to be removed in one of the subsequent patches */
- if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs")) {
- strcpy(lsi->lsi_fstype, "ldiskfs");
- } else {
- strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
- }
-
- /* Determine server type */
- rc = server_name2index(lsi->lsi_svname, &index, NULL);
- if (rc < 0) {
- if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
- /* Assume we're a bare MGS */
- rc = 0;
- lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
- } else {
- LCONSOLE_ERROR("Can't determine server type of '%s'\n",
- lsi->lsi_svname);
- RETURN(rc);
- }
- }
- lsi->lsi_flags |= rc;
-
- /* Add mount line flags that used to be in ldd:
- * writeconf, mgs, anything else?
- */
- lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
- LDD_F_WRITECONF : 0;
- lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
- LDD_F_VIRGIN : 0;
- lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
- LDD_F_SV_TYPE_MGS : 0;
- lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
- LDD_F_NO_PRIMNODE : 0;
-
- RETURN(0);
-}
-
-/*************** server mount ******************/
-
-/** Start the shutdown of servers at umount.
- */
-static void server_put_super(struct super_block *sb)
-{
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct obd_device *obd;
- char *tmpname, *extraname = NULL;
- int tmpname_sz;
- int lsiflags = lsi->lsi_flags;
- ENTRY;
-
- LASSERT(IS_SERVER(lsi));
-
- tmpname_sz = strlen(lsi->lsi_svname) + 1;
- OBD_ALLOC(tmpname, tmpname_sz);
- memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
- CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
- if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
- snprintf(tmpname, tmpname_sz, "MGS");
-
- /* disconnect the osp-on-ost first to drain off the inflight request */
- if (IS_OST(lsi) || IS_MDT(lsi)) {
- int rc;
-
- rc = lustre_disconnect_osp(sb);
- if (rc && rc != ETIMEDOUT)
- CERROR("%s: failed to disconnect osp-on-ost (rc=%d)!\n",
- tmpname, rc);
- }
-
- /* Stop the target */
- if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
- (IS_MDT(lsi) || IS_OST(lsi))) {
- struct lustre_profile *lprof = NULL;
-
- /* tell the mgc to drop the config log */
- lustre_end_log(sb, lsi->lsi_svname, NULL);
-
- /* COMPAT_146 - profile may get deleted in mgc_cleanup.
- If there are any setup/cleanup errors, save the lov
- name for safety cleanup later. */
- lprof = class_get_profile(lsi->lsi_svname);
- if (lprof && lprof->lp_dt) {
- OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
- strcpy(extraname, lprof->lp_dt);
- }
-
- obd = class_name2obd(lsi->lsi_svname);
- if (obd) {
- CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
- if (lsiflags & LSI_UMOUNT_FAILOVER)
- obd->obd_fail = 1;
- /* We can't seem to give an error return code
- * to .put_super, so we better make sure we clean up! */
- obd->obd_force = 1;
- class_manual_cleanup(obd);
- } else {
- CERROR("no obd %s\n", lsi->lsi_svname);
- server_deregister_mount(lsi->lsi_svname);
- }
- }
-
- /* If they wanted the mgs to stop separately from the mdt, they
- should have put it on a different device. */
- if (IS_MGS(lsi)) {
- /* if MDS start with --nomgs, don't stop MGS then */
- if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
- server_stop_mgs(sb);
- }
-
- if (IS_OST(lsi) || IS_MDT(lsi)) {
- if (lustre_stop_osp(sb) < 0)
- CERROR("%s: Fail to stop osp-on-ost!\n", tmpname);
- }
-
- /* Clean the mgc and sb */
- lustre_common_put_super(sb);
-
- /* wait till all in-progress cleanups are done
- * specifically we're interested in ofd cleanup
- * as it pins OSS */
- obd_zombie_barrier();
-
- /* Stop the servers (MDS, OSS) if no longer needed. We must wait
- until the target is really gone so that our type refcount check
- is right. */
- server_stop_servers(lsiflags);
-
- /* In case of startup or cleanup err, stop related obds */
- if (extraname) {
- obd = class_name2obd(extraname);
- if (obd) {
- CWARN("Cleaning orphaned obd %s\n", extraname);
- obd->obd_force = 1;
- class_manual_cleanup(obd);
- }
- OBD_FREE(extraname, strlen(extraname) + 1);
- }
-
- LCONSOLE_WARN("server umount %s complete\n", tmpname);
- OBD_FREE(tmpname, tmpname_sz);
- EXIT;
-}
-
-/** Called only for 'umount -f'
- */
-#ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
-static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
-{
- struct super_block *sb = vfsmnt->mnt_sb;
-#else
-static void server_umount_begin(struct super_block *sb)
-{
-#endif
- struct lustre_sb_info *lsi = s2lsi(sb);
- ENTRY;
-
-#ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
- if (!(flags & MNT_FORCE)) {
- EXIT;
- return;
- }
-#endif
-
- CDEBUG(D_MOUNT, "umount -f\n");
- /* umount = failover
- umount -f = force
- no third way to do non-force, non-failover */
- lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
- EXIT;
-}
-
-static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
-{
- struct super_block *sb = dentry->d_sb;
- struct lustre_sb_info *lsi = s2lsi(sb);
- struct obd_statfs statfs;
- int rc;
- ENTRY;
-
- if (lsi->lsi_dt_dev) {
- rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
- if (rc == 0) {
- statfs_unpack(buf, &statfs);
- buf->f_type = sb->s_magic;
- RETURN(0);
- }
- }
-
- /* just return 0 */
- buf->f_type = sb->s_magic;
- buf->f_bsize = sb->s_blocksize;
- buf->f_blocks = 1;
- buf->f_bfree = 0;
- buf->f_bavail = 0;
- buf->f_files = 1;
- buf->f_ffree = 0;
- buf->f_namelen = NAME_MAX;
- RETURN(0);
-}
-
-/** The operations we support directly on the superblock:
- * mount, umount, and df.
- */
-static struct super_operations server_ops =
-{
- .put_super = server_put_super,
- .umount_begin = server_umount_begin, /* umount -f */
- .statfs = server_statfs,
-};
-
-#define log2(n) ffz(~(n))
-#define LUSTRE_SUPER_MAGIC 0x0BD00BD1
-
-static int server_fill_super_common(struct super_block *sb)
-{
- struct inode *root = 0;
- ENTRY;
-
- CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
-
- sb->s_blocksize = 4096;
- sb->s_blocksize_bits = log2(sb->s_blocksize);
- sb->s_magic = LUSTRE_SUPER_MAGIC;
- sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
- sb->s_flags |= MS_RDONLY;
- sb->s_op = &server_ops;
-
- root = new_inode(sb);
- if (!root) {
- CERROR("Can't make root inode\n");
- RETURN(-EIO);
- }
-
- /* returns -EIO for every operation */
- /* make_bad_inode(root); -- badness - can't umount */
- /* apparently we need to be a directory for the mount to finish */
- root->i_mode = S_IFDIR;
-
- sb->s_root = d_make_root(root);
- if (!sb->s_root) {
- CERROR("%s: can't make root dentry\n", sb->s_id);
- RETURN(-EIO);
- }
-
- RETURN(0);
-}
-
-static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
-{
- struct lustre_mount_data *lmd = lsi->lsi_lmd;
- struct obd_device *obd;
- struct dt_device_param p;
- char flagstr[16];
- int rc;
- ENTRY;
-
- CDEBUG(D_MOUNT,
- "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
- lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
-
- sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
- strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
- strcat(lsi->lsi_osd_uuid, "_UUID");
- sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
-
- obd = class_name2obd(lsi->lsi_osd_obdname);
- if (obd == NULL) {
- rc = lustre_start_simple(lsi->lsi_osd_obdname,
- lsi->lsi_osd_type,
- lsi->lsi_osd_uuid, lmd->lmd_dev,
- flagstr, lsi->lsi_lmd->lmd_opts,
- lsi->lsi_svname);
- if (rc)
- GOTO(out, rc);
- obd = class_name2obd(lsi->lsi_osd_obdname);
- LASSERT(obd);
- }
-
- rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL);
- if (rc) {
- obd->obd_force = 1;
- class_manual_cleanup(obd);
- lsi->lsi_dt_dev = NULL;
- }
-
- /* XXX: to keep support old components relying on lsi_srv_mnt
- * we get this info from OSD just started */
- LASSERT(obd->obd_lu_dev);
- lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
- LASSERT(lsi->lsi_dt_dev);
-
- dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
-
- lsi->lsi_srv_mnt = p.ddp_mnt;
-
-out:
- RETURN(rc);
-}
-
-/** Fill in the superblock info for a Lustre server.
- * Mount the device with the correct options.
- * Read the on-disk config file.
- * Start the services.
- */
-static int server_fill_super(struct super_block *sb)
-{
- struct lustre_sb_info *lsi = s2lsi(sb);
- int rc;
- ENTRY;
-
- rc = lsi_prepare(lsi);
- if (rc)
- RETURN(rc);
-
- /* Start low level OSD */
- rc = osd_start(lsi, sb->s_flags);
- if (rc) {
- CERROR("Unable to start osd on %s: %d\n",
- lsi->lsi_lmd->lmd_dev, rc);
- lustre_put_lsi(sb);
- RETURN(rc);
- }
-
- CDEBUG(D_MOUNT, "Found service %s on device %s\n",
- lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
-
- if (class_name2obd(lsi->lsi_svname)) {
- LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
- "running. Double-mount may have compromised"
- " the disk journal.\n",
- lsi->lsi_svname);
- lustre_put_lsi(sb);
- RETURN(-EALREADY);
- }
-
- /* Start MGS before MGC */
- if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
- rc = server_start_mgs(sb);
- if (rc)
- GOTO(out_mnt, rc);
- }
-
- /* Start MGC before servers */
- rc = lustre_start_mgc(sb);
- if (rc)
- GOTO(out_mnt, rc);
-
- /* Set up all obd devices for service */
- if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
- (IS_OST(lsi) || IS_MDT(lsi))) {
- rc = server_start_targets(sb, lsi->lsi_srv_mnt);
- if (rc < 0) {
- CERROR("Unable to start targets: %d\n", rc);
- GOTO(out_mnt, rc);
- }
- /* FIXME overmount client here,
- or can we just start a client log and client_fill_super on this sb?
- We need to make sure server_put_super gets called too - ll_put_super
- calls lustre_common_put_super; check there for LSI_SERVER flag,
- call s_p_s if so.
- Probably should start client from new thread so we can return.
- Client will not finish until all servers are connected.
- Note - MGS-only server does NOT get a client, since there is no
- lustre fs associated - the MGS is for all lustre fs's */
- }
-
- rc = server_fill_super_common(sb);
- if (rc)
- GOTO(out_mnt, rc);
-
- RETURN(0);
-out_mnt:
- /* We jump here in case of failure while starting targets or MGS.
- * In this case we can't just put @mnt and have to do real cleanup
- * with stoping targets, etc. */
- server_put_super(sb);
- return rc;
-}
-
-/*
- * Calculate timeout value for a target.
- */
-void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
-{
- struct lustre_mount_data *lmd;
- int soft = 0;
- int hard = 0;
- int factor = 0;
- bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
- int min = OBD_RECOVERY_TIME_MIN;
-
- LASSERT(IS_SERVER(lsi));
-
- lmd = lsi->lsi_lmd;
- if (lmd) {
- soft = lmd->lmd_recovery_time_soft;
- hard = lmd->lmd_recovery_time_hard;
- has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
- obd->obd_no_ir = !has_ir;
- }
-
- if (soft == 0)
- soft = OBD_RECOVERY_TIME_SOFT;
- if (hard == 0)
- hard = OBD_RECOVERY_TIME_HARD;
-
- /* target may have ir_factor configured. */
- factor = OBD_IR_FACTOR_DEFAULT;
- if (obd->obd_recovery_ir_factor)
- factor = obd->obd_recovery_ir_factor;
-
- if (has_ir) {
- int new_soft = soft;
- int new_hard = hard;
-
- /* adjust timeout value by imperative recovery */
-
- new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
- new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
-
- /* make sure the timeout is not too short */
- new_soft = max(min, new_soft);
- new_hard = max(new_soft, new_hard);