}
};
+static int ofd_stack_init(const struct lu_env *env,
+ struct ofd_device *m, struct lustre_cfg *cfg)
+{
+ struct lu_device *ofd_lu = &m->ofd_dt_dev.dd_lu_dev;
+ const char *dev = lustre_cfg_string(cfg, 0);
+ struct obd_type *type;
+ struct lu_device_type *ldt;
+ struct lu_device *d;
+ struct ofd_thread_info *info = ofd_info(env);
+ int rc;
+
+ ENTRY;
+
+ /* XXX: we should be able to use different OSDs here */
+ type = class_get_type(LUSTRE_OSD_NAME);
+ if (!type) {
+ CERROR("Unknown type: '%s'\n", LUSTRE_OSD_NAME);
+ RETURN(-ENODEV);
+ }
+
+ rc = lu_env_refill((struct lu_env *)env);
+ if (rc != 0) {
+ CERROR("Failure to refill session: '%d'\n", rc);
+ GOTO(out_type, rc);
+ }
+
+ ldt = type->typ_lu;
+ if (ldt == NULL) {
+ CERROR("type: '%s'\n", LUSTRE_OSD_NAME);
+ GOTO(out_type, rc = -EINVAL);
+ }
+
+ ldt->ldt_obd_type = type;
+ d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
+ if (IS_ERR(d)) {
+ CERROR("Cannot allocate device: '%s'\n", LUSTRE_OSD_NAME);
+ GOTO(out_type, rc = -ENODEV);
+ }
+
+ LASSERT(ofd_lu->ld_site);
+ d->ld_site = ofd_lu->ld_site;
+
+ snprintf(info->fti_u.name, sizeof(info->fti_u.name),
+ "%s-osd", lustre_cfg_string(cfg, 0));
+
+ type->typ_refcnt++;
+ rc = ldt->ldt_ops->ldto_device_init(env, d, dev, NULL);
+ if (rc) {
+ CERROR("can't init device '%s', rc %d\n", LUSTRE_OSD_NAME, rc);
+ GOTO(out_free, rc);
+ }
+ lu_device_get(d);
+ lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
+
+ m->ofd_osd = lu2dt_dev(d);
+
+ /* process setup config */
+ rc = d->ld_ops->ldo_process_config(env, d, cfg);
+ if (rc)
+ GOTO(out_fini, rc);
+
+ RETURN(rc);
+
+out_fini:
+ ldt->ldt_ops->ldto_device_fini(env, d);
+out_free:
+ type->typ_refcnt--;
+ ldt->ldt_ops->ldto_device_free(env, d);
+out_type:
+ class_put_type(type);
+ RETURN(rc);
+}
+
+static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
+ struct lu_device *top)
+{
+ struct obd_device *obd = ofd_obd(m);
+ struct lustre_cfg_bufs bufs;
+ struct lustre_cfg *lcfg;
+ char flags[3] = "";
+
+ ENTRY;
+
+ lu_site_purge(env, top->ld_site, ~0);
+
+ /* process cleanup, pass mdt obd name to get obd umount flags */
+ lustre_cfg_bufs_reset(&bufs, obd->obd_name);
+ if (obd->obd_force)
+ strcat(flags, "F");
+ if (obd->obd_fail)
+ strcat(flags, "A");
+ lustre_cfg_bufs_set_string(&bufs, 1, flags);
+ lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
+ if (!lcfg) {
+ CERROR("Cannot alloc lcfg!\n");
+ RETURN_EXIT;
+ }
+
+ LASSERT(top);
+ top->ld_ops->ldo_process_config(env, top, lcfg);
+ lustre_cfg_free(lcfg);
+
+ lu_stack_fini(env, &m->ofd_osd->dd_lu_dev);
+ m->ofd_osd = NULL;
+
+ EXIT;
+}
+
/* used by MGS to process specific configurations */
static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
struct lustre_cfg *cfg)
{
- int rc = 0;
+ struct ofd_device *m = ofd_dev(d);
+ struct dt_device *dt_next = m->ofd_osd;
+ struct lu_device *next = &dt_next->dd_lu_dev;
+ int rc;
ENTRY;
lprocfs_ofd_init_vars(&lvars);
rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, cfg,
d->ld_obd);
+ if (rc > 0 || rc == -ENOSYS)
+ /* we don't understand; pass it on */
+ rc = next->ld_ops->ldo_process_config(env, next, cfg);
break;
}
case LCFG_SPTLRPC_CONF: {
break;
}
default:
+ /* others are passed further */
+ rc = next->ld_ops->ldo_process_config(env, next, cfg);
break;
}
RETURN(rc);
}
+static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
+ const struct lu_object_conf *conf)
+{
+ struct ofd_device *d = ofd_dev(o->lo_dev);
+ struct lu_device *under;
+ struct lu_object *below;
+ int rc = 0;
+
+ ENTRY;
+
+ CDEBUG(D_INFO, "object init, fid = "DFID"\n",
+ PFID(lu_object_fid(o)));
+
+ under = &d->ofd_osd->dd_lu_dev;
+ below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
+ if (below != NULL)
+ lu_object_add(o, below);
+ else
+ rc = -ENOMEM;
+
+ RETURN(rc);
+}
+
+static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
+{
+ struct ofd_object *of = ofd_obj(o);
+ struct lu_object_header *h;
+
+ ENTRY;
+
+ h = o->lo_header;
+ CDEBUG(D_INFO, "object free, fid = "DFID"\n",
+ PFID(lu_object_fid(o)));
+
+ lu_object_fini(o);
+ lu_object_header_fini(h);
+ OBD_SLAB_FREE_PTR(of, ofd_object_kmem);
+ EXIT;
+}
+
+static int ofd_object_print(const struct lu_env *env, void *cookie,
+ lu_printer_t p, const struct lu_object *o)
+{
+ return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
+}
+
struct lu_object_operations ofd_obj_ops = {
+ .loo_object_init = ofd_object_init,
+ .loo_object_free = ofd_object_free,
+ .loo_object_print = ofd_object_print
};
static struct lu_object *ofd_object_alloc(const struct lu_env *env,
}
}
-static int ofd_start(const struct lu_env *env, struct lu_device *parent,
- struct lu_device *dev)
+extern int ost_handle(struct ptlrpc_request *req);
+
+static int ofd_start(const struct lu_env *env, struct lu_device *dev)
{
- struct obd_device *obd = dev->ld_obd;
- int rc = 0;
+ struct ofd_device *ofd = ofd_dev(dev);
+ struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
+ int rc;
ENTRY;
- LASSERT(obd->obd_no_conn);
- cfs_spin_lock(&obd->obd_dev_lock);
- obd->obd_no_conn = 0;
- cfs_spin_unlock(&obd->obd_dev_lock);
+ /* initialize lower device */
+ rc = next->ld_ops->ldo_prepare(env, dev, next);
RETURN(rc);
}
static int ofd_recovery_complete(const struct lu_env *env,
struct lu_device *dev)
{
- int rc = 0;
+ struct ofd_device *ofd = ofd_dev(dev);
+ struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
+ int rc = 0;
ENTRY;
+ /* Grant space for object precreation on the self export.
+ * This initial reserved space (i.e. 20MB for zfs and 560KB for ldiskfs)
+ * is enough to create 20k objects. It is then adapted based on the
+ * precreate request size (see ofd_grant_create()
+ */
+ ofd_grant_connect(env, dev->ld_obd->obd_self_export,
+ OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace);
+ rc = next->ld_ops->ldo_recovery_complete(env, next);
RETURN(rc);
}
static struct lu_device_operations ofd_lu_ops = {
.ldo_object_alloc = ofd_object_alloc,
.ldo_process_config = ofd_process_config,
- .ldo_prepare = ofd_start,
.ldo_recovery_complete = ofd_recovery_complete,
};
+static int ofd_procfs_init(struct ofd_device *ofd)
+{
+ struct lprocfs_static_vars lvars;
+ struct obd_device *obd = ofd_obd(ofd);
+ cfs_proc_dir_entry_t *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 */
+ lprocfs_ofd_init_vars(&lvars);
+ rc = lprocfs_obd_setup(obd, lvars.obd_vars);
+ if (rc) {
+ CERROR("%s: lprocfs_obd_setup failed: %d.\n",
+ obd->obd_name, rc);
+ RETURN(rc);
+ }
+
+ rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_LAST);
+ if (rc) {
+ CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
+ obd->obd_name, rc);
+ GOTO(obd_cleanup, rc);
+ }
+
+ /* Init OFD private stats here */
+ lprocfs_counter_init(obd->obd_stats, LPROC_OFD_READ_BYTES,
+ LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
+ lprocfs_counter_init(obd->obd_stats, LPROC_OFD_WRITE_BYTES,
+ LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
+
+ rc = lproc_ofd_attach_seqstat(obd);
+ if (rc) {
+ CERROR("%s: create seqstat failed: %d.\n", obd->obd_name, rc);
+ GOTO(free_obd_stats, 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(free_obd_stats, rc);
+ }
+ obd->obd_proc_exports_entry = entry;
+
+ entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
+ lprocfs_nid_stats_clear_read,
+ lprocfs_nid_stats_clear_write, obd, NULL);
+ if (IS_ERR(entry)) {
+ rc = PTR_ERR(entry);
+ CERROR("%s: add proc entry 'clear' failed: %d.\n",
+ obd->obd_name, rc);
+ GOTO(free_obd_stats, rc);
+ }
+ RETURN(0);
+
+free_obd_stats:
+ lprocfs_free_obd_stats(obd);
+obd_cleanup:
+ lprocfs_obd_cleanup(obd);
+ return rc;
+}
+
+static int ofd_procfs_fini(struct ofd_device *ofd)
+{
+ struct obd_device *obd = ofd_obd(ofd);
+
+ lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
+ lprocfs_free_per_client_stats(obd);
+ lprocfs_free_obd_stats(obd);
+ lprocfs_obd_cleanup(obd);
+ return 0;
+}
+
+extern int ost_handle(struct ptlrpc_request *req);
+
static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
struct lu_device_type *ldt, struct lustre_cfg *cfg)
{
const char *dev = lustre_cfg_string(cfg, 0);
+ struct ofd_thread_info *info = NULL;
struct obd_device *obd;
+ struct obd_statfs *osfs;
int rc;
ENTRY;
obd->u.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;
+
+ cfs_spin_lock_init(&m->ofd_flags_lock);
+ m->ofd_raid_degraded = 0;
+ m->ofd_syncjournal = 0;
+ ofd_slc_set(m);
+ m->ofd_grant_compat_disable = 0;
+
+ /* statfs data */
+ cfs_spin_lock_init(&m->ofd_osfs_lock);
+ m->ofd_osfs_age = cfs_time_shift_64(-1000);
+ m->ofd_osfs_unstable = 0;
+ m->ofd_statfs_inflight = 0;
+ m->ofd_osfs_inflight = 0;
+
+ /* grant data */
+ cfs_spin_lock_init(&m->ofd_grant_lock);
+ m->ofd_tot_dirty = 0;
+ m->ofd_tot_granted = 0;
+ m->ofd_tot_pending = 0;
+ m->ofd_max_group = 0;
+
+ cfs_rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
+ sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
+
+ obd->u.filter.fo_fl_oss_capa = 0;
+ CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
+ obd->u.filter.fo_capa_hash = init_capa_hash();
+ if (obd->u.filter.fo_capa_hash == NULL)
+ RETURN(-ENOMEM);
+
m->ofd_dt_dev.dd_lu_dev.ld_ops = &ofd_lu_ops;
m->ofd_dt_dev.dd_lu_dev.ld_obd = obd;
/* 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 */
obd->obd_no_conn = 1;
obd->obd_replayable = 1;
}
}
+ info = ofd_info_init(env, NULL);
+ if (info == NULL)
+ RETURN(-EFAULT);
+
+ rc = lu_site_init(&m->ofd_site, &m->ofd_dt_dev.dd_lu_dev);
+ if (rc)
+ GOTO(err_fini_proc, rc);
+ m->ofd_site.ls_top_dev = &m->ofd_dt_dev.dd_lu_dev;
+
+ rc = ofd_stack_init(env, m, cfg);
+ if (rc) {
+ CERROR("Can't init device stack, rc %d\n", rc);
+ GOTO(err_lu_site, rc);
+ }
+
+ /* populate cached statfs data */
+ osfs = &ofd_info(env)->fti_u.osfs;
+ rc = ofd_statfs_internal(env, m, osfs, 0, NULL);
+ if (rc != 0) {
+ CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc);
+ GOTO(err_fini_stack, rc);
+ }
+ if (!IS_PO2(osfs->os_bsize)) {
+ CERROR("%s: blocksize (%d) is not a power of 2\n",
+ obd->obd_name, osfs->os_bsize);
+ GOTO(err_fini_stack, rc = -EPROTO);
+ }
+ m->ofd_blockbits = cfs_fls(osfs->os_bsize) - 1;
+
+ snprintf(info->fti_u.name, sizeof(info->fti_u.name), "filter-%p", m);
+ m->ofd_namespace = ldlm_namespace_new(obd, info->fti_u.name,
+ LDLM_NAMESPACE_SERVER,
+ LDLM_NAMESPACE_GREEDY,
+ LDLM_NS_TYPE_OST);
+ if (m->ofd_namespace == NULL)
+ GOTO(err_fini_stack, rc = -ENOMEM);
+ /* set obd_namespace for compatibility with old code */
+ obd->obd_namespace = m->ofd_namespace;
+
+ dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf);
+
+ /* Allow at most ddp_grant_reserved% of the available filesystem space
+ * to be granted to clients, so that any errors in the grant overhead
+ * calculations do not allow granting more space to clients than can be
+ * written. Assumes that in aggregate the grant overhead calculations do
+ * not have more than ddp_grant_reserved% estimation error in them. */
+ m->ofd_grant_ratio =
+ ofd_grant_ratio_conv(m->ofd_dt_conf.ddp_grant_reserved);
+
+ rc = ofd_start(env, &m->ofd_dt_dev.dd_lu_dev);
+ if (rc)
+ GOTO(err_fini_stack, rc);
+
+ rc = lut_init(env, &m->ofd_lut, obd, m->ofd_osd);
+ if (rc)
+ GOTO(err_free_ns, rc);
+
+ rc = ofd_fs_setup(env, m, obd);
+ if (rc)
+ GOTO(err_fini_lut, rc);
+
+ target_recovery_init(&m->ofd_lut, ost_handle);
+
+ rc = lu_site_init_finish(&m->ofd_site);
+ if (rc)
+ GOTO(err_fs_cleanup, rc);
+
RETURN(0);
+err_fs_cleanup:
+ target_recovery_fini(obd);
+ ofd_fs_cleanup(env, m);
+err_fini_lut:
+ lut_fini(env, &m->ofd_lut);
+err_free_ns:
+ ldlm_namespace_free(m->ofd_namespace, 0, obd->obd_force);
+ obd->obd_namespace = m->ofd_namespace = NULL;
+err_fini_stack:
+ ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev);
+err_lu_site:
+ lu_site_fini(&m->ofd_site);
+err_fini_proc:
+ ofd_procfs_fini(m);
+ return rc;
}
static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
struct obd_device *obd = ofd_obd(m);
struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
+ target_recovery_fini(obd);
obd_exports_barrier(obd);
obd_zombie_barrier();
+ lut_fini(env, &m->ofd_lut);
+ ofd_fs_cleanup(env, m);
+
+ ofd_free_capa_keys(m);
+ cleanup_capa_hash(obd->u.filter.fo_capa_hash);
+
+ if (m->ofd_namespace != NULL) {
+ ldlm_namespace_free(m->ofd_namespace, NULL,
+ d->ld_obd->obd_force);
+ d->ld_obd->obd_namespace = m->ofd_namespace = NULL;
+ }
+
+ ofd_stack_fini(env, m, m->ofd_site.ls_top_dev);
+ lu_site_fini(&m->ofd_site);
+ ofd_procfs_fini(m);
LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
EXIT;
}
-static struct lu_device* ofd_device_fini(const struct lu_env *env,
+static struct lu_device *ofd_device_fini(const struct lu_env *env,
struct lu_device *d)
{
ENTRY;
struct ofd_thread_info *info = data;
info->fti_env = NULL;
+ info->fti_exp = NULL;
}
struct lu_context_key ofd_thread_key = {
if (rc)
return rc;
+ rc = ofd_fmd_init();
+ if (rc) {
+ lu_kmem_fini(ofd_caches);
+ return(rc);
+ }
+
lprocfs_ofd_init_vars(&lvars);
rc = class_register_type(&ofd_obd_ops, NULL, lvars.module_vars,
void __exit ofd_exit(void)
{
- class_unregister_type(LUSTRE_OST_NAME);
+ ofd_fmd_exit();
lu_kmem_fini(ofd_caches);
+ class_unregister_type(LUSTRE_OST_NAME);
}
MODULE_AUTHOR("Whamcloud, Inc. <http://www.whamcloud.com/>");
git://git.whamcloud.com / fs / lustre-release.git / blobdiff