-
- 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;
- int cplen = 0;
- ENTRY;
-
- if (!IS_SERVER(lsi))
- RETURN(-EINVAL);
-
- if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
- >= sizeof(mti->mti_svname))
- RETURN(-E2BIG);
-
- 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;
- cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
- sizeof(mti->mti_params));
- if (cplen >= sizeof(mti->mti_params))
- return -E2BIG;
- 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 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_lwp(sb);
- if (rc) {
- CERROR("%s: failed to start LWP: %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 lwp first to drain off the inflight request */
- if (IS_OST(lsi) || IS_MDT(lsi)) {
- int rc;
-
- rc = lustre_disconnect_lwp(sb);
- if (rc && rc != ETIMEDOUT)
- CERROR("%s: failed to disconnect lwp. (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_lwp(sb) < 0)
- CERROR("%s: failed to stop lwp!\n", tmpname);