X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fofd%2Fofd_dev.c;h=843abca369bcf40976ae12fd2260151c7ec74ea1;hp=7495453948aea2845b7dd4a7df5fd5c44e925d90;hb=3cce65712d94cffe8f1626545845b95b88aef672;hpb=711eff3a50003b5c617955bae50fba7bfcb9c43d diff --git a/lustre/ofd/ofd_dev.c b/lustre/ofd/ofd_dev.c index 7495453..843abca 100644 --- a/lustre/ofd/ofd_dev.c +++ b/lustre/ofd/ofd_dev.c @@ -82,7 +82,6 @@ /* Slab for OFD object allocation */ static struct kmem_cache *ofd_object_kmem; - static struct lu_kmem_descr ofd_caches[] = { { .ckd_cache = &ofd_object_kmem, @@ -320,13 +319,7 @@ static bool match_symlink_param(char *param) sval = strchr(param, '='); if (sval != NULL) { paramlen = sval - param; - if (strncmp(param, "writethrough_cache_enable", - paramlen) == 0 || - strncmp(param, "readcache_max_filesize", - paramlen) == 0 || - strncmp(param, "read_cache_enable", - paramlen) == 0 || - strncmp(param, "brw_stats", paramlen) == 0) + if (strncmp(param, "brw_stats", paramlen) == 0) return true; } } @@ -359,11 +352,11 @@ static int ofd_process_config(const struct lu_env *env, struct lu_device *d, switch (cfg->lcfg_command) { case LCFG_PARAM: { - struct obd_device *obd = ofd_obd(m); /* For interoperability */ - struct cfg_interop_param *ptr = NULL; - struct lustre_cfg *old_cfg = NULL; - char *param = NULL; + struct cfg_interop_param *ptr = NULL; + struct lustre_cfg *old_cfg = NULL; + char *param = NULL; + ssize_t count; param = lustre_cfg_string(cfg, 1); if (param == NULL) { @@ -397,14 +390,16 @@ static int ofd_process_config(const struct lu_env *env, struct lu_device *d, break; } - rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg, - d->ld_obd); - if (rc > 0 || rc == -ENOSYS) { - CDEBUG(D_CONFIG, "pass param %s down the stack.\n", - param); - /* we don't understand; pass it on */ - rc = next->ld_ops->ldo_process_config(env, next, cfg); + count = class_modify_config(cfg, PARAM_OST, + &d->ld_obd->obd_kset.kobj); + if (count > 0) { + rc = 0; + break; } + CDEBUG(D_CONFIG, "pass param %s down the stack.\n", + param); + /* we don't understand; pass it on */ + rc = next->ld_ops->ldo_process_config(env, next, cfg); break; } case LCFG_SPTLRPC_CONF: { @@ -638,6 +633,8 @@ static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev, LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc); target_recovery_init(&ofd->ofd_lut, tgt_request_handle); + OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE, + (OBD_TIMEOUT_DEFAULT + 1) / 4); LASSERT(obd->obd_no_conn); spin_lock(&obd->obd_dev_lock); obd->obd_no_conn = 0; @@ -699,73 +696,7 @@ static struct lu_device_operations ofd_lu_ops = { .ldo_prepare = ofd_prepare, }; -LPROC_SEQ_FOPS(lprocfs_nid_stats_clear); - -/** - * Initialize all needed procfs entries for OFD device. - * - * \param[in] ofd OFD device - * - * \retval 0 if successful - * \retval negative value on error - */ -static int ofd_procfs_init(struct ofd_device *ofd) -{ - struct obd_device *obd = ofd_obd(ofd); - struct proc_dir_entry *entry; - int rc = 0; - - ENTRY; - - /* lprocfs must be setup before the ofd so state can be safely added - * to /proc incrementally as the ofd is setup */ - obd->obd_vars = lprocfs_ofd_obd_vars; - rc = lprocfs_obd_setup(obd, false); - if (rc) { - CERROR("%s: lprocfs_obd_setup failed: %d.\n", - obd->obd_name, rc); - RETURN(rc); - } - - rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST); - if (rc) { - CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n", - obd->obd_name, rc); - GOTO(obd_cleanup, rc); - } - - entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL); - if (IS_ERR(entry)) { - rc = PTR_ERR(entry); - CERROR("%s: error %d setting up lprocfs for %s\n", - obd->obd_name, rc, "exports"); - GOTO(obd_cleanup, rc); - } - obd->obd_proc_exports_entry = entry; - - entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear", - obd, &lprocfs_nid_stats_clear_fops); - if (IS_ERR(entry)) { - rc = PTR_ERR(entry); - CERROR("%s: add proc entry 'clear' failed: %d.\n", - obd->obd_name, rc); - GOTO(obd_cleanup, rc); - } - - ofd_stats_counter_init(obd->obd_stats); - - rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST, - ofd_stats_counter_init); - if (rc) - GOTO(obd_cleanup, rc); - RETURN(0); -obd_cleanup: - lprocfs_obd_cleanup(obd); - lprocfs_free_obd_stats(obd); - - return rc; -} - +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0) /** * Expose OSD statistics to OFD layer. * @@ -779,8 +710,31 @@ obd_cleanup: */ static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd) { - struct obd_device *obd = ofd_obd(ofd); - struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd; + struct obd_device *obd = ofd_obd(ofd); + struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd; + struct kobj_type *osd_type; + int i; + + osd_type = get_ktype(&ofd->ofd_osd->dd_kobj); + for (i = 0; osd_type->default_attrs[i]; i++) { + if (strcmp(osd_type->default_attrs[i]->name, + "read_cache_enable") == 0) { + ofd->ofd_read_cache_enable = + osd_type->default_attrs[i]; + } + + if (strcmp(osd_type->default_attrs[i]->name, + "readcache_max_filesize") == 0) { + ofd->ofd_read_cache_max_filesize = + osd_type->default_attrs[i]; + } + + if (strcmp(osd_type->default_attrs[i]->name, + "writethrough_cache_enable") == 0) { + ofd->ofd_write_cache_enable = + osd_type->default_attrs[i]; + } + } if (obd->obd_proc_entry == NULL) return; @@ -788,21 +742,8 @@ static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd) lprocfs_add_symlink("brw_stats", obd->obd_proc_entry, "../../%s/%s/brw_stats", osd_obd->obd_type->typ_name, obd->obd_name); - - lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry, - "../../%s/%s/read_cache_enable", - osd_obd->obd_type->typ_name, obd->obd_name); - - lprocfs_add_symlink("readcache_max_filesize", - obd->obd_proc_entry, - "../../%s/%s/readcache_max_filesize", - osd_obd->obd_type->typ_name, obd->obd_name); - - lprocfs_add_symlink("writethrough_cache_enable", - obd->obd_proc_entry, - "../../%s/%s/writethrough_cache_enable", - osd_obd->obd_type->typ_name, obd->obd_name); } +#endif /** * Cleanup all procfs entries in OFD. @@ -813,6 +754,7 @@ static void ofd_procfs_fini(struct ofd_device *ofd) { struct obd_device *obd = ofd_obd(ofd); + tgt_tunables_fini(&ofd->ofd_lut); lprocfs_free_per_client_stats(obd); lprocfs_obd_cleanup(obd); lprocfs_free_obd_stats(obd); @@ -852,18 +794,19 @@ int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd) */ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) { - struct seq_server_site *ss = &ofd->ofd_seq_site; - struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev; - char *obd_name = ofd_name(ofd); - char *name = NULL; - int rc = 0; + struct seq_server_site *ss = &ofd->ofd_seq_site; + struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev; + char *obd_name = ofd_name(ofd); + char *name = NULL; + int len = strlen(obd_name) + 7; + int rc = 0; ss = &ofd->ofd_seq_site; lu->ld_site->ld_seq_site = ss; ss->ss_lu = lu->ld_site; ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index; - OBD_ALLOC(name, sizeof(obd_name) * 2 + 10); + OBD_ALLOC(name, len); if (name == NULL) return -ENOMEM; @@ -874,7 +817,7 @@ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name, LUSTRE_SEQ_SERVER, ss); if (rc) { - CERROR("%s : seq server init error %d\n", obd_name, rc); + CERROR("%s: seq server init error: rc = %d\n", obd_name, rc); GOTO(out_server, rc); } ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id; @@ -883,15 +826,11 @@ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) if (ss->ss_client_seq == NULL) GOTO(out_server, rc = -ENOMEM); - /* - * It always printed as "%p", so that the name is unique in the kernel, - * even if the filesystem is mounted twice. So sizeof(.) * 2 is enough. - */ - snprintf(name, sizeof(obd_name) * 2 + 7, "%p-super", obd_name); + snprintf(name, len, "%s-super", obd_name); rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA, name, NULL); if (rc) { - CERROR("%s : seq client init error %d\n", obd_name, rc); + CERROR("%s: seq client init error: rc = %d\n", obd_name, rc); GOTO(out_client, rc); } @@ -908,7 +847,7 @@ out_server: ss->ss_server_seq = NULL; } out_name: - OBD_FREE(name, sizeof(obd_name) * 2 + 10); + OBD_FREE(name, len); return rc; } @@ -1021,7 +960,8 @@ int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd, } -static int ofd_lock_unlock_region(struct ldlm_namespace *ns, +static int ofd_lock_unlock_region(const struct lu_env *env, + struct ldlm_namespace *ns, struct ldlm_res_id *res_id, unsigned long long begin, unsigned long long end) @@ -1032,7 +972,7 @@ static int ofd_lock_unlock_region(struct ldlm_namespace *ns, LASSERT(begin <= end); - rc = tgt_extent_lock(ns, res_id, begin, end, &lh, LCK_PR, &flags); + rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags); if (rc != 0) return rc; @@ -1060,7 +1000,8 @@ static int ofd_lock_unlock_region(struct ldlm_namespace *ns, * \retval 0 if successful * \retval negative value on error */ -static int lock_zero_regions(struct ldlm_namespace *ns, +static int lock_zero_regions(const struct lu_env *env, + struct ldlm_namespace *ns, struct ldlm_res_id *res_id, struct fiemap *fiemap) { @@ -1076,7 +1017,7 @@ static int lock_zero_regions(struct ldlm_namespace *ns, if (fiemap_start[i].fe_logical > begin) { CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n", begin, fiemap_start[i].fe_logical); - rc = ofd_lock_unlock_region(ns, res_id, begin, + rc = ofd_lock_unlock_region(env, ns, res_id, begin, fiemap_start[i].fe_logical); if (rc) RETURN(rc); @@ -1088,7 +1029,7 @@ static int lock_zero_regions(struct ldlm_namespace *ns, if (begin < (fiemap->fm_start + fiemap->fm_length)) { CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n", begin, fiemap->fm_start + fiemap->fm_length); - rc = ofd_lock_unlock_region(ns, res_id, begin, + rc = ofd_lock_unlock_region(env, ns, res_id, begin, fiemap->fm_start + fiemap->fm_length); } @@ -1189,7 +1130,7 @@ static int ofd_get_info_hdl(struct tgt_session_info *tsi) if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS && fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) { ost_fid_build_resid(fid, &fti->fti_resid); - rc = lock_zero_regions(ofd->ofd_namespace, + rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace, &fti->fti_resid, fiemap); if (rc == 0) rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, @@ -1281,7 +1222,8 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH)) lock_mode = LCK_PW; - rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace, + rc = tgt_extent_lock(tsi->tsi_env, + tsi->tsi_tgt->lut_obd->obd_namespace, &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh, lock_mode, &flags); if (rc != 0) @@ -1600,8 +1542,9 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) oseq->os_destroys_in_progress = 1; mutex_lock(&oseq->os_create_lock); if (!oseq->os_destroys_in_progress) { - CERROR("%s:[%llu] destroys_in_progress already" - " cleared\n", ofd_name(ofd), seq); + CDEBUG(D_HA, + "%s:[%llu] destroys_in_progress already cleared\n", + ofd_name(ofd), seq); rc = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq)); GOTO(out, rc); @@ -1680,6 +1623,7 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) u64 next_id; int created = 0; int count; + int rc2; if (!(oa->o_valid & OBD_MD_FLFLAGS) || !(oa->o_flags & OBD_FL_DELORPHAN)) { @@ -1767,11 +1711,14 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) granted = 0; } - rc = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq)); + rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq)); + rc = rc ? : rc2; } EXIT; ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE, tsi->tsi_jobid, 1); + if (unlikely(!oseq->os_last_id_synced)) + oseq->os_last_id_synced = 1; out: mutex_unlock(&oseq->os_create_lock); out_nolock: @@ -2037,8 +1984,8 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi) oa->o_flags & OBD_FL_SRVLOCK; if (srvlock) { - rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh, - LCK_PW, &flags); + rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start, + end, &lh, LCK_PW, &flags); if (rc != 0) RETURN(rc); } @@ -2248,7 +2195,7 @@ static int ofd_ladvise_hdl(struct tgt_session_info *tsi) ioo.ioo_oid = body->oa.o_oi; ioo.ioo_bufcnt = 1; - rc = tgt_extent_lock(exp->exp_obd->obd_namespace, + rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace, &tsi->tsi_resid, start, end - 1, &lockh, LCK_PR, &flags); if (rc != 0) @@ -2764,26 +2711,25 @@ TGT_RPC_HANDLER(OST_FIRST_OPC, 0, OST_SET_INFO, ofd_set_info_hdl, &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION), TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl), -TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl), -TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_SETATTR, ofd_setattr_hdl), -TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR, +TGT_OST_HDL(HAS_REPLY | IS_MUTABLE, OST_CREATE, ofd_create_hdl), -TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR, +TGT_OST_HDL(HAS_REPLY | IS_MUTABLE, OST_DESTROY, ofd_destroy_hdl), -TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl), -TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO, - OST_BRW_READ, tgt_brw_read, +TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl), +TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read, ofd_hp_brw), /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */ -TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write, +TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write, ofd_hp_brw), -TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR, +TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_PUNCH, ofd_punch_hdl, ofd_hp_punch), -TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl), -TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl), -TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl), +TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl), }; static struct tgt_opc_slice ofd_common_slice[] = { @@ -2905,13 +2851,10 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, obt = &obd->u.obt; obt->obt_magic = OBT_MAGIC; - m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT; - m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT; - spin_lock_init(&m->ofd_flags_lock); m->ofd_raid_degraded = 0; m->ofd_checksum_t10pi_enforce = 0; - m->ofd_syncjournal = 0; + m->ofd_sync_journal = 0; ofd_slc_set(m); m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT; @@ -2928,12 +2871,6 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, /* set this lu_device to obd, because error handling need it */ obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev; - rc = ofd_procfs_init(m); - if (rc) { - CERROR("Can't init ofd lprocfs, rc %d\n", rc); - RETURN(rc); - } - /* No connection accepted until configurations will finish */ spin_lock(&obd->obd_dev_lock); obd->obd_no_conn = 1; @@ -2950,15 +2887,18 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, info = ofd_info_init(env, NULL); if (info == NULL) - GOTO(err_fini_proc, rc = -EFAULT); + RETURN(-EFAULT); rc = ofd_stack_init(env, m, cfg); if (rc) { - CERROR("Can't init device stack, rc %d\n", rc); - GOTO(err_fini_proc, rc); + CERROR("%s: can't init device stack, rc %d\n", + obd->obd_name, rc); + RETURN(rc); } +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0) ofd_procfs_add_brw_stats_symlink(m); +#endif snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s", "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid); @@ -2983,6 +2923,10 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, if (rc) GOTO(err_free_ns, rc); + rc = ofd_tunables_init(m); + if (rc) + GOTO(err_fini_lut, rc); + tgd->tgd_reserved_pcnt = 0; m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size; @@ -2995,7 +2939,7 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, rc = ofd_fs_setup(env, m, obd); if (rc) - GOTO(err_fini_lut, rc); + GOTO(err_fini_proc, rc); fid.f_seq = FID_SEQ_LOCAL_NAME; fid.f_oid = 1; @@ -3031,6 +2975,8 @@ err_fini_los: m->ofd_los = NULL; err_fini_fs: ofd_fs_cleanup(env, m); +err_fini_proc: + ofd_procfs_fini(m); err_fini_lut: tgt_fini(env, &m->ofd_lut); err_free_ns: @@ -3038,8 +2984,6 @@ err_free_ns: obd->obd_namespace = m->ofd_namespace = NULL; err_fini_stack: ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev); -err_fini_proc: - ofd_procfs_fini(m); return rc; } @@ -3070,6 +3014,7 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m) obd_exports_barrier(obd); obd_zombie_barrier(); + ofd_procfs_fini(m); tgt_fini(env, &m->ofd_lut); ofd_stop_inconsistency_verification_thread(m); lfsck_degister(env, m->ofd_osd); @@ -3083,7 +3028,7 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m) } ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev); - ofd_procfs_fini(m); + LASSERT(atomic_read(&d->ld_ref) == 0); server_put_mount(obd->obd_name, true); EXIT; @@ -3202,13 +3147,6 @@ static int __init ofd_init(void) rc = lu_kmem_init(ofd_caches); if (rc) return rc; - - rc = ofd_fmd_init(); - if (rc) { - lu_kmem_fini(ofd_caches); - return(rc); - } - rc = class_register_type(&ofd_obd_ops, NULL, true, NULL, LUSTRE_OST_NAME, &ofd_device_type); return rc; @@ -3222,7 +3160,6 @@ static int __init ofd_init(void) */ static void __exit ofd_exit(void) { - ofd_fmd_exit(); lu_kmem_fini(ofd_caches); class_unregister_type(LUSTRE_OST_NAME); }