* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Intel, Inc.
+ * Copyright (c) 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <obd_class.h>
#include <lustre_param.h>
+#include <lustre_log.h>
#include "osp_internal.h"
if (rc)
GOTO(out, rc);
}
+ CDEBUG(D_HA, "%s: Read last used ID: "LPU64"\n", m->opd_obd->obd_name,
+ le64_to_cpu(m->opd_last_used_id));
RETURN(0);
out:
CERROR("%s: can't initialize lov_objid: %d\n",
static void osp_last_used_fini(const struct lu_env *env, struct osp_device *d)
{
- lu_object_put(env, &d->opd_last_used_file->do_lu);
- d->opd_last_used_file = NULL;
+ if (d->opd_last_used_file != NULL) {
+ lu_object_put(env, &d->opd_last_used_file->do_lu);
+ d->opd_last_used_file = NULL;
+ }
}
-static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
+int osp_disconnect(struct osp_device *d)
{
- struct obd_import *imp;
- int rc = 0;
- ENTRY;
-
- /* release last_used file */
- osp_last_used_fini(env, d);
+ struct obd_import *imp;
+ int rc = 0;
imp = d->opd_obd->u.cli.cl_import;
/* Mark import deactivated now, so we don't try to reconnect if any
* of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
* fully deactivate the import, or that would drop all requests. */
- cfs_spin_lock(&imp->imp_lock);
+ LASSERT(imp != NULL);
+ spin_lock(&imp->imp_lock);
imp->imp_deactive = 1;
- cfs_spin_unlock(&imp->imp_lock);
+ spin_unlock(&imp->imp_lock);
ptlrpc_deactivate_import(imp);
(void)ptlrpc_pinger_del_import(imp);
rc = ptlrpc_disconnect_import(imp, 0);
+ if (rc == -ETIMEDOUT || rc == -ENOTCONN || rc == -ESHUTDOWN)
+ rc = 0;
if (rc)
CERROR("%s: can't disconnect: rc = %d\n",
d->opd_obd->obd_name, rc);
ptlrpc_invalidate_import(imp);
+ RETURN(rc);
+}
+
+static int osp_shutdown(const struct lu_env *env, struct osp_device *d)
+{
+ int rc = 0;
+ ENTRY;
+
+ if (is_osp_on_ost(d->opd_obd->obd_name)) {
+ rc = osp_disconnect(d);
+ RETURN(rc);
+ }
+
+ LASSERT(env);
+ /* release last_used file */
+ osp_last_used_fini(env, d);
+
+ rc = osp_disconnect(d);
+
/* stop precreate thread */
osp_precreate_fini(d);
+ /* stop sync thread */
+ osp_sync_fini(d);
+
RETURN(rc);
}
switch (lcfg->lcfg_command) {
case LCFG_CLEANUP:
- lu_dev_del_linkage(dev->ld_site, dev);
+ if (!is_osp_on_ost(d->opd_obd->obd_name))
+ lu_dev_del_linkage(dev->ld_site, dev);
rc = osp_shutdown(env, d);
break;
case LCFG_PARAM:
lprocfs_osp_init_vars(&lvars);
LASSERT(d->opd_obd);
- rc = class_process_proc_param(PARAM_OSP, lvars.obd_vars,
+ rc = class_process_proc_param(PARAM_OSC, lvars.obd_vars,
lcfg, d->opd_obd);
if (rc > 0)
rc = 0;
.ldo_recovery_complete = osp_recovery_complete,
};
+/**
+ * provides with statfs from corresponded OST
+ *
+ */
+static int osp_statfs(const struct lu_env *env, struct dt_device *dev,
+ struct obd_statfs *sfs)
+{
+ struct osp_device *d = dt2osp_dev(dev);
+
+ ENTRY;
+
+ if (unlikely(d->opd_imp_active == 0))
+ RETURN(-ENOTCONN);
+
+ /* return recently updated data */
+ *sfs = d->opd_statfs;
+
+ /*
+ * layer above osp (usually lod) can use ffree to estimate
+ * how many objects are available for immediate creation
+ */
+ spin_lock(&d->opd_pre_lock);
+ sfs->os_fprecreated = d->opd_pre_last_created - d->opd_pre_used_id;
+ sfs->os_fprecreated -= d->opd_pre_reserved;
+ spin_unlock(&d->opd_pre_lock);
+
+ LASSERT(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2);
+
+ CDEBUG(D_OTHER, "%s: "LPU64" blocks, "LPU64" free, "LPU64" avail, "
+ LPU64" files, "LPU64" free files\n", d->opd_obd->obd_name,
+ sfs->os_blocks, sfs->os_bfree, sfs->os_bavail,
+ sfs->os_files, sfs->os_ffree);
+ RETURN(0);
+}
+
static int osp_sync(const struct lu_env *env, struct dt_device *dev)
{
ENTRY;
RETURN(0);
}
-static const struct dt_device_operations osp_dt_ops = {
+const struct dt_device_operations osp_dt_ops = {
+ .dt_statfs = osp_statfs,
.dt_sync = osp_sync,
};
static int osp_init0(const struct lu_env *env, struct osp_device *m,
struct lu_device_type *ldt, struct lustre_cfg *cfg)
{
- struct lprocfs_static_vars lvars = { 0 };
- struct proc_dir_entry *osc_proc_dir;
+ struct obd_device *obd;
struct obd_import *imp;
class_uuid_t uuid;
char *src, *ost, *mdt, *osdname = NULL;
ENTRY;
- m->opd_obd = class_name2obd(lustre_cfg_string(cfg, 0));
- if (m->opd_obd == NULL) {
+ obd = class_name2obd(lustre_cfg_string(cfg, 0));
+ if (obd == NULL) {
CERROR("Cannot find obd with name %s\n",
lustre_cfg_string(cfg, 0));
RETURN(-ENODEV);
}
+ m->opd_obd = obd;
/* There is no record in the MDT configuration for the local disk
* device, so we have to extract this from elsewhere in the profile.
idx = simple_strtol(ost + 4, &mdt, 16);
if (mdt[0] != '-' || idx > INT_MAX || idx < 0) {
- CERROR("%s: invalid OST index in '%s'\n",
- m->opd_obd->obd_name, src);
+ CERROR("%s: invalid OST index in '%s'\n", obd->obd_name, src);
GOTO(out_fini, rc = -EINVAL);
}
m->opd_index = idx;
idx = ost - src;
/* check the fsname length, and after this everything else will fit */
if (idx > MTI_NAME_MAXLEN) {
- CERROR("%s: fsname too long in '%s'\n",
- m->opd_obd->obd_name, src);
+ CERROR("%s: fsname too long in '%s'\n", obd->obd_name, src);
GOTO(out_fini, rc = -EINVAL);
}
else
strcat(osdname, mdt);
strcat(osdname, "-osd");
- CDEBUG(D_HA, "%s: connect to %s (%s)\n",
- m->opd_obd->obd_name, osdname, src);
+ CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src);
m->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops;
m->opd_dt_dev.dd_ops = &osp_dt_ops;
- m->opd_obd->obd_lu_dev = &m->opd_dt_dev.dd_lu_dev;
+ obd->obd_lu_dev = &m->opd_dt_dev.dd_lu_dev;
rc = osp_connect_to_osd(env, m, osdname);
if (rc)
if (rc)
GOTO(out_disconnect, rc);
- rc = client_obd_setup(m->opd_obd, cfg);
+ rc = client_obd_setup(obd, cfg);
if (rc) {
CERROR("%s: can't setup obd: %d\n", m->opd_obd->obd_name, rc);
GOTO(out_ref, rc);
}
- lprocfs_osp_init_vars(&lvars);
- if (lprocfs_obd_setup(m->opd_obd, lvars.obd_vars) == 0)
- ptlrpc_lprocfs_register_obd(m->opd_obd);
-
- /* for compatibility we link old procfs's OSC entries to osp ones */
- osc_proc_dir = lprocfs_srch(proc_lustre_root, "osc");
- if (osc_proc_dir) {
- cfs_proc_dir_entry_t *symlink = NULL;
- char *name;
-
- OBD_ALLOC(name, strlen(m->opd_obd->obd_name) + 1);
- if (name == NULL)
- GOTO(out, rc = -ENOMEM);
-
- strcpy(name, m->opd_obd->obd_name);
- if (strstr(name, "osc"))
- symlink = lprocfs_add_symlink(name, osc_proc_dir,
- "../osp/%s",
- m->opd_obd->obd_name);
- OBD_FREE(name, strlen(m->opd_obd->obd_name) + 1);
- m->opd_symlink = symlink;
- }
+ osp_lprocfs_init(m);
/*
* Initialize last id from the storage - will be used in orphan cleanup
GOTO(out_last_used, rc);
/*
+ * Initialize synhronization mechanism taking care of propogating
+ * changes to OST in near transactional manner
+ */
+ rc = osp_sync_init(env, m);
+ if (rc)
+ GOTO(out_precreat, rc);
+
+ /*
* Initiate connect to OST
*/
ll_generate_random_uuid(uuid);
class_uuid_unparse(uuid, &m->opd_cluuid);
- imp = m->opd_obd->u.cli.cl_import;
+ imp = obd->u.cli.cl_import;
rc = ptlrpc_init_import(imp);
if (rc)
RETURN(0);
out:
+ /* stop sync thread */
+ osp_sync_fini(m);
+out_precreat:
/* stop precreate thread */
osp_precreate_fini(m);
out_last_used:
osp_last_used_fini(env, m);
out_proc:
- ptlrpc_lprocfs_unregister_obd(m->opd_obd);
- lprocfs_obd_cleanup(m->opd_obd);
- class_destroy_import(m->opd_obd->u.cli.cl_import);
- client_obd_cleanup(m->opd_obd);
+ ptlrpc_lprocfs_unregister_obd(obd);
+ lprocfs_obd_cleanup(obd);
+ class_destroy_import(obd->u.cli.cl_import);
+ client_obd_cleanup(obd);
out_ref:
ptlrpcd_decref();
out_disconnect:
ENTRY;
+ if (cfs_atomic_read(&lu->ld_ref) && lu->ld_site) {
+ LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
+ lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
+ }
dt_device_fini(&m->opd_dt_dev);
OBD_FREE_PTR(m);
RETURN(NULL);
l = osp2lu_dev(m);
dt_device_init(&m->opd_dt_dev, t);
- rc = osp_init0(env, m, t, lcfg);
+ if (is_osp_on_ost(lustre_cfg_string(lcfg, 0)))
+ rc = osp_init_for_ost(env, m, t, lcfg);
+ else
+ rc = osp_init0(env, m, t, lcfg);
if (rc != 0) {
osp_device_free(env, l);
l = ERR_PTR(rc);
ENTRY;
- LASSERT(m->opd_storage_exp);
- obd_disconnect(m->opd_storage_exp);
+ if (m->opd_storage_exp)
+ obd_disconnect(m->opd_storage_exp);
+
+ if (is_osp_on_ost(m->opd_obd->obd_name))
+ osp_fini_for_ost(m);
imp = m->opd_obd->u.cli.cl_import;
RETURN(rc);
*exp = class_conn2export(&conn);
+ if (is_osp_on_ost(obd->obd_name))
+ osp->opd_exp = *exp;
/* Why should there ever be more than 1 connect? */
osp->opd_connects++;
OBD_CONNECT_OSS_CAPA |
OBD_CONNECT_REQPORTAL |
OBD_CONNECT_SKIP_ORPHAN |
- OBD_CONNECT_VERSION;
+ OBD_CONNECT_VERSION |
+ OBD_CONNECT_FID |
+ OBD_CONNECT_LVB_TYPE;
+
+ if (is_osp_on_ost(osp->opd_obd->obd_name))
+ ocd->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
+
ocd->ocd_version = LUSTRE_VERSION_CODE;
LASSERT(data->ocd_connect_flags & OBD_CONNECT_INDEX);
ocd->ocd_index = data->ocd_index;
struct obd_device *obd = exp->exp_obd;
struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
int rc;
-
ENTRY;
/* Only disconnect the underlying layers on the final disconnect. */
osp->opd_connects--;
rc = class_disconnect(exp);
+ if (rc) {
+ CERROR("%s: class disconnect error: rc = %d\n",
+ obd->obd_name, rc);
+ RETURN(rc);
+ }
/* destroy the device */
- if (rc == 0)
+ if (!is_osp_on_ost(obd->obd_name))
class_manual_cleanup(obd);
RETURN(rc);
}
+/*
+ * lprocfs helpers still use OBD API, let's keep obd_statfs() support
+ */
+static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp,
+ struct obd_statfs *osfs, __u64 max_age, __u32 flags)
+{
+ struct obd_statfs *msfs;
+ struct ptlrpc_request *req;
+ struct obd_import *imp = NULL;
+ int rc;
+
+ ENTRY;
+
+ /* Since the request might also come from lprocfs, so we need
+ * sync this with client_disconnect_export Bug15684 */
+ down_read(&exp->exp_obd->u.cli.cl_sem);
+ if (exp->exp_obd->u.cli.cl_import)
+ imp = class_import_get(exp->exp_obd->u.cli.cl_import);
+ up_read(&exp->exp_obd->u.cli.cl_sem);
+ if (!imp)
+ RETURN(-ENODEV);
+
+ /* We could possibly pass max_age in the request (as an absolute
+ * timestamp or a "seconds.usec ago") so the target can avoid doing
+ * extra calls into the filesystem if that isn't necessary (e.g.
+ * during mount that would help a bit). Having relative timestamps
+ * is not so great if request processing is slow, while absolute
+ * timestamps are not ideal because they need time synchronization. */
+ req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS);
+
+ class_import_put(imp);
+
+ if (req == NULL)
+ RETURN(-ENOMEM);
+
+ rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS);
+ if (rc) {
+ ptlrpc_request_free(req);
+ RETURN(rc);
+ }
+ ptlrpc_request_set_replen(req);
+ req->rq_request_portal = OST_CREATE_PORTAL;
+ ptlrpc_at_set_req_timeout(req);
+
+ if (flags & OBD_STATFS_NODELAY) {
+ /* procfs requests not want stat in wait for avoid deadlock */
+ req->rq_no_resend = 1;
+ req->rq_no_delay = 1;
+ }
+
+ rc = ptlrpc_queue_wait(req);
+ if (rc)
+ GOTO(out, rc);
+
+ msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
+ if (msfs == NULL)
+ GOTO(out, rc = -EPROTO);
+
+ *osfs = *msfs;
+
+ EXIT;
+out:
+ ptlrpc_req_finished(req);
+ return rc;
+}
+
+static int osp_import_event(struct obd_device *obd, struct obd_import *imp,
+ enum obd_import_event event)
+{
+ struct osp_device *d = lu2osp_dev(obd->obd_lu_dev);
+
+ switch (event) {
+ case IMP_EVENT_DISCON:
+ d->opd_got_disconnected = 1;
+ d->opd_imp_connected = 0;
+ if (is_osp_on_ost(d->opd_obd->obd_name))
+ break;
+ osp_pre_update_status(d, -ENODEV);
+ cfs_waitq_signal(&d->opd_pre_waitq);
+ CDEBUG(D_HA, "got disconnected\n");
+ break;
+ case IMP_EVENT_INACTIVE:
+ d->opd_imp_active = 0;
+ if (is_osp_on_ost(d->opd_obd->obd_name))
+ break;
+ osp_pre_update_status(d, -ENODEV);
+ cfs_waitq_signal(&d->opd_pre_waitq);
+ CDEBUG(D_HA, "got inactive\n");
+ break;
+ case IMP_EVENT_ACTIVE:
+ d->opd_imp_active = 1;
+ if (d->opd_got_disconnected)
+ d->opd_new_connection = 1;
+ d->opd_imp_connected = 1;
+ d->opd_imp_seen_connected = 1;
+ if (is_osp_on_ost(d->opd_obd->obd_name))
+ break;
+ cfs_waitq_signal(&d->opd_pre_waitq);
+ __osp_sync_check_for_work(d);
+ CDEBUG(D_HA, "got connected\n");
+ break;
+ case IMP_EVENT_INVALIDATE:
+ if (obd->obd_namespace == NULL)
+ break;
+ ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
+ break;
+ case IMP_EVENT_OCD:
+ case IMP_EVENT_DEACTIVATE:
+ case IMP_EVENT_ACTIVATE:
+ break;
+ default:
+ CERROR("%s: unsupported import event: %#x\n",
+ obd->obd_name, event);
+ }
+ return 0;
+}
+
+static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
+ void *karg, void *uarg)
+{
+ struct obd_device *obd = exp->exp_obd;
+ struct osp_device *d;
+ struct obd_ioctl_data *data = karg;
+ int rc = 0;
+
+ ENTRY;
+
+ LASSERT(obd->obd_lu_dev);
+ d = lu2osp_dev(obd->obd_lu_dev);
+ LASSERT(d->opd_dt_dev.dd_ops == &osp_dt_ops);
+
+ if (!cfs_try_module_get(THIS_MODULE)) {
+ CERROR("%s: can't get module. Is it alive?", obd->obd_name);
+ return -EINVAL;
+ }
+
+ switch (cmd) {
+ case OBD_IOC_CLIENT_RECOVER:
+ rc = ptlrpc_recover_import(obd->u.cli.cl_import,
+ data->ioc_inlbuf1, 0);
+ if (rc > 0)
+ rc = 0;
+ break;
+ case IOC_OSC_SET_ACTIVE:
+ rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
+ data->ioc_offset);
+ break;
+ case OBD_IOC_PING_TARGET:
+ rc = ptlrpc_obd_ping(obd);
+ break;
+ default:
+ CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name,
+ cmd, cfs_curproc_comm());
+ rc = -ENOTTY;
+ }
+ cfs_module_put(THIS_MODULE);
+ return rc;
+}
+
+static int osp_obd_health_check(const struct lu_env *env,
+ struct obd_device *obd)
+{
+ struct osp_device *d = lu2osp_dev(obd->obd_lu_dev);
+
+ ENTRY;
+
+ /*
+ * 1.8/2.0 behaviour is that OST being connected once at least
+ * is considired "healthy". and one "healty" OST is enough to
+ * allow lustre clients to connect to MDS
+ */
+ LASSERT(d);
+ RETURN(!d->opd_imp_seen_connected);
+}
+
/* context key constructor/destructor: mdt_key_init, mdt_key_fini */
LU_KEY_INIT_FINI(osp, struct osp_thread_info);
static void osp_key_exit(const struct lu_context *ctx,
.o_reconnect = osp_reconnect,
.o_connect = osp_obd_connect,
.o_disconnect = osp_obd_disconnect,
+ .o_health_check = osp_obd_health_check,
+ .o_import_event = osp_import_event,
+ .o_iocontrol = osp_iocontrol,
+ .o_statfs = osp_obd_statfs,
};
+struct llog_operations osp_mds_ost_orig_logops;
+
static int __init osp_mod_init(void)
{
struct lprocfs_static_vars lvars;
return rc;
}
+ /* Note: add_rec/delcare_add_rec will be only used by catalogs */
+ osp_mds_ost_orig_logops = llog_osd_ops;
+ osp_mds_ost_orig_logops.lop_add = llog_cat_add_rec;
+ osp_mds_ost_orig_logops.lop_declare_add = llog_cat_declare_add_rec;
+
osc_proc_dir = lprocfs_srch(proc_lustre_root, "osc");
if (osc_proc_dir == NULL) {
osc_proc_dir = lprocfs_register("osc", proc_lustre_root, NULL,
MODULE_LICENSE("GPL");
cfs_module(osp, LUSTRE_VERSION_STRING, osp_mod_init, osp_mod_exit);
-