* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_MDS
-#include <obd_class.h>
+#include <linux/kthread.h>
+
#include <lustre_ioctl.h>
-#include <lustre_param.h>
#include <lustre_log.h>
-#include <lustre_mdc.h>
+#include <lustre_obdo.h>
+#include <lustre_param.h>
+#include <obd_class.h>
#include "osp_internal.h"
* \retval object object being created if the creation succeed.
* \retval NULL NULL if the creation failed.
*/
-struct lu_object *osp_object_alloc(const struct lu_env *env,
- const struct lu_object_header *hdr,
- struct lu_device *d)
+static struct lu_object *osp_object_alloc(const struct lu_env *env,
+ const struct lu_object_header *hdr,
+ struct lu_device *d)
{
struct lu_object_header *h = NULL;
struct osp_object *o;
RETURN(dto);
/* Get attributes of the local object. */
- rc = dt_attr_get(env, dto, attr, NULL);
+ rc = dt_attr_get(env, dto, attr);
if (rc) {
CERROR("%s: can't be initialized: rc = %d\n",
osp->opd_obd->obd_name, rc);
struct osp_thread_info *osi = osp_env_info(env);
struct lu_fid *fid = &osp->opd_last_used_fid;
struct dt_object *dto;
- int rc;
+ int rc = -EFAULT;
ENTRY;
dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
osp->opd_index);
rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
- if (rc != 0)
+ if (rc != 0 && rc != -EFAULT)
GOTO(out, rc);
- } else {
+ }
+
+ if (rc == -EFAULT) { /* fresh LAST_ID */
fid->f_oid = 0;
osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid,
osp->opd_index);
struct osp_thread_info *osi = osp_env_info(env);
struct lu_fid *fid = &osp->opd_last_used_fid;
struct dt_object *dto;
- int rc;
+ int rc = -EFAULT;
ENTRY;
dto = osp_find_or_create_local_file(env, osp, &osi->osi_attr,
osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
osp->opd_index);
rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off);
- if (rc != 0)
+ if (rc != 0 && rc != -EFAULT)
GOTO(out, rc);
- } else {
+ }
+
+ if (rc == -EFAULT) { /* fresh OSP */
fid->f_seq = 0;
osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq,
osp->opd_index);
rc = osp_write_local_file(env, osp, dto, &osi->osi_lb,
osi->osi_off);
+ if (rc != 0)
+ GOTO(out, rc);
}
osp->opd_last_used_seq_file = dto;
RETURN(0);
}
/**
+ * Initialize the osp_update structure in OSP device
+ *
+ * Allocate osp update structure and start update thread.
+ *
+ * \param[in] osp OSP device
+ *
+ * \retval 0 if initialization succeeds.
+ * \retval negative errno if initialization fails.
+ */
+static int osp_update_init(struct osp_device *osp)
+{
+ struct l_wait_info lwi = { 0 };
+ struct task_struct *task;
+
+ ENTRY;
+
+ LASSERT(osp->opd_connect_mdt);
+
+ OBD_ALLOC_PTR(osp->opd_update);
+ if (osp->opd_update == NULL)
+ RETURN(-ENOMEM);
+
+ init_waitqueue_head(&osp->opd_update_thread.t_ctl_waitq);
+ init_waitqueue_head(&osp->opd_update->ou_waitq);
+ spin_lock_init(&osp->opd_update->ou_lock);
+ INIT_LIST_HEAD(&osp->opd_update->ou_list);
+ osp->opd_update->ou_rpc_version = 1;
+ osp->opd_update->ou_version = 1;
+
+ /* start thread handling sending updates to the remote MDT */
+ task = kthread_run(osp_send_update_thread, osp,
+ "osp_up%u-%u", osp->opd_index, osp->opd_group);
+ if (IS_ERR(task)) {
+ int rc = PTR_ERR(task);
+
+ OBD_FREE_PTR(osp->opd_update);
+ osp->opd_update = NULL;
+ CERROR("%s: can't start precreate thread: rc = %d\n",
+ osp->opd_obd->obd_name, rc);
+ RETURN(rc);
+ }
+
+ l_wait_event(osp->opd_update_thread.t_ctl_waitq,
+ osp_send_update_thread_running(osp) ||
+ osp_send_update_thread_stopped(osp), &lwi);
+
+ RETURN(0);
+}
+
+/**
+ * Finialize osp_update structure in OSP device
+ *
+ * Stop the OSP update sending thread, then delete the left
+ * osp thandle in the sending list.
+ *
+ * \param [in] osp OSP device.
+ */
+static void osp_update_fini(const struct lu_env *env, struct osp_device *osp)
+{
+ struct osp_update_request *our;
+ struct osp_update_request *tmp;
+ struct osp_updates *ou = osp->opd_update;
+
+ if (ou == NULL)
+ return;
+
+ osp->opd_update_thread.t_flags = SVC_STOPPING;
+ wake_up(&ou->ou_waitq);
+
+ wait_event(osp->opd_update_thread.t_ctl_waitq,
+ osp->opd_update_thread.t_flags & SVC_STOPPED);
+
+ /* Remove the left osp thandle from the list */
+ spin_lock(&ou->ou_lock);
+ list_for_each_entry_safe(our, tmp, &ou->ou_list,
+ our_list) {
+ list_del_init(&our->our_list);
+ LASSERT(our->our_th != NULL);
+ osp_trans_callback(env, our->our_th, -EIO);
+ /* our will be destroyed in osp_thandle_put() */
+ osp_thandle_put(our->our_th);
+ }
+ spin_unlock(&ou->ou_lock);
+
+ OBD_FREE_PTR(ou);
+ osp->opd_update = NULL;
+}
+
+/**
* Cleanup OSP, which includes disconnect import, cleanup unlink log, stop
* precreate threads etc.
*
rc = osp_disconnect(d);
- osp_sync_fini(d);
-
if (!d->opd_connect_mdt) {
+ /* stop sync thread */
+ osp_sync_fini(d);
+
/* stop precreate thread */
osp_precreate_fini(d);
switch (lcfg->lcfg_command) {
case LCFG_PRE_CLEANUP:
rc = osp_disconnect(d);
+ osp_update_fini(env, d);
break;
case LCFG_CLEANUP:
lu_dev_del_linkage(dev->ld_site, dev);
break;
case LCFG_PARAM:
LASSERT(obd);
- rc = class_process_proc_param(PARAM_OSC, obd->obd_vars,
- lcfg, obd);
+ rc = class_process_proc_param(d->opd_connect_mdt ?
+ PARAM_OSP : PARAM_OSC,
+ obd->obd_vars, lcfg, obd);
if (rc > 0)
rc = 0;
if (rc == -ENOSYS) {
struct obd_statfs *sfs)
{
struct osp_device *d = dt2osp_dev(dev);
+ struct obd_import *imp = d->opd_obd->u.cli.cl_import;
ENTRY;
+ if (imp->imp_state == LUSTRE_IMP_CLOSED)
+ RETURN(-ESHUTDOWN);
+
if (unlikely(d->opd_imp_active == 0))
RETURN(-ENOTCONN);
unsigned long start = cfs_time_current();
ENTRY;
+ /* No Sync between MDTs yet. */
+ if (d->opd_connect_mdt)
+ RETURN(0);
+
if (unlikely(d->opd_imp_active == 0))
RETURN(-ENOTCONN);
id = d->opd_syn_last_used_id;
+ down_write(&d->opd_async_updates_rwsem);
+
+ CDEBUG(D_OTHER, "%s: async updates %d\n", d->opd_obd->obd_name,
+ atomic_read(&d->opd_async_updates_count));
+
+ /* make sure the connection is fine */
+ expire = cfs_time_shift(obd_timeout);
+ lwi = LWI_TIMEOUT(expire - cfs_time_current(), osp_sync_timeout, d);
+ rc = l_wait_event(d->opd_syn_barrier_waitq,
+ atomic_read(&d->opd_async_updates_count) == 0,
+ &lwi);
+ up_write(&d->opd_async_updates_rwsem);
+ if (rc != 0)
+ GOTO(out, rc);
- CDEBUG(D_OTHER, "%s: id: used %lu, processed %lu\n",
+ CDEBUG(D_CACHE, "%s: id: used %lu, processed %lu\n",
d->opd_obd->obd_name, id, d->opd_syn_last_processed_id);
/* wait till all-in-line are processed */
/* block new processing (barrier>0 - few callers are possible */
atomic_inc(&d->opd_syn_barrier);
- CDEBUG(D_OTHER, "%s: %u in flight\n", d->opd_obd->obd_name,
+ CDEBUG(D_CACHE, "%s: %u in flight\n", d->opd_obd->obd_name,
d->opd_syn_rpc_in_flight);
/* wait till all-in-flight are replied, so executed by the target */
GOTO(out, rc = -ETIMEDOUT);
}
- CDEBUG(D_OTHER, "%s: done in %lu\n", d->opd_obd->obd_name,
- cfs_time_current() - start);
out:
/* resume normal processing (barrier=0) */
atomic_dec(&d->opd_syn_barrier);
__osp_sync_check_for_work(d);
+ CDEBUG(D_CACHE, "%s: done in %lu: rc = %d\n", d->opd_obd->obd_name,
+ cfs_time_current() - start, rc);
+
RETURN(rc);
}
.dt_trans_create = osp_trans_create,
.dt_trans_start = osp_trans_start,
.dt_trans_stop = osp_trans_stop,
+ .dt_trans_cb_add = osp_trans_cb_add,
};
/**
}
/**
+ * Determine if the lock needs to be cancelled
+ *
+ * Determine if the unused lock should be cancelled before replay, see
+ * (ldlm_cancel_no_wait_policy()). Currently, only inode bits lock exists
+ * between MDTs.
+ *
+ * \param[in] lock lock to be checked.
+ *
+ * \retval 1 if the lock needs to be cancelled before replay.
+ * \retval 0 if the lock does not need to be cancelled before
+ * replay.
+ */
+static int osp_cancel_weight(struct ldlm_lock *lock)
+{
+ if (lock->l_resource->lr_type != LDLM_IBITS)
+ RETURN(0);
+
+ RETURN(1);
+}
+
+/**
* Initialize OSP device according to the parameters in the configuration
* log \a cfg.
*
ENTRY;
mutex_init(&osp->opd_async_requests_mutex);
+ INIT_LIST_HEAD(&osp->opd_async_updates);
+ init_rwsem(&osp->opd_async_updates_rwsem);
+ atomic_set(&osp->opd_async_updates_count, 0);
obd = class_name2obd(lustre_cfg_string(cfg, 0));
if (obd == NULL) {
* used in orphan cleanup. */
rc = osp_last_used_init(env, osp);
if (rc)
- GOTO(out_proc, rc);
+ GOTO(out_fid, rc);
/* Initialize precreation thread, it handles new
rc = osp_init_precreate(osp);
if (rc)
GOTO(out_last_used, rc);
+
+ /*
+ * Initialize synhronization mechanism taking
+ * care of propogating changes to OST in near
+ * transactional manner.
+ */
+ rc = osp_sync_init(env, osp);
+ if (rc < 0)
+ GOTO(out_precreat, rc);
+ } else {
+ rc = osp_update_init(osp);
+ if (rc != 0)
+ GOTO(out_fid, rc);
}
- /*
- * Initialize synhronization mechanism taking
- * care of propogating changes to OST in near
- * transactional manner.
- */
- rc = osp_sync_init(env, osp);
- if (rc)
- GOTO(out_precreat, rc);
+ ns_register_cancel(obd->obd_namespace, osp_cancel_weight);
/*
* Initiate connect to OST
RETURN(0);
out:
- /* stop sync thread */
- osp_sync_fini(osp);
+ if (!osp->opd_connect_mdt)
+ /* stop sync thread */
+ osp_sync_fini(osp);
out_precreat:
/* stop precreate thread */
if (!osp->opd_connect_mdt)
osp_precreate_fini(osp);
+ else
+ osp_update_fini(env, osp);
out_last_used:
if (!osp->opd_connect_mdt)
osp_last_used_fini(env, osp);
+out_fid:
+ obd_fid_fini(osp->opd_obd);
out_proc:
ptlrpc_lprocfs_unregister_obd(obd);
lprocfs_obd_cleanup(obd);
ENTRY;
if (osp->opd_async_requests != NULL) {
- out_destroy_update_req(osp->opd_async_requests);
+ osp_update_request_destroy(osp->opd_async_requests);
osp->opd_async_requests = NULL;
}
imp = osp->opd_obd->u.cli.cl_import;
- if (imp->imp_rq_pool) {
- ptlrpc_free_rq_pool(imp->imp_rq_pool);
- imp->imp_rq_pool = NULL;
- }
-
if (osp->opd_symlink)
lprocfs_remove(&osp->opd_symlink);
if (rc) {
CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc);
GOTO(out, rc);
+ } else {
+ osp->opd_obd->u.cli.cl_seq->lcs_exp =
+ class_export_get(osp->opd_exp);
}
ptlrpc_pinger_add_import(imp);
}
/**
- * Prepare fid client.
- *
- * This function prepares the FID client for the OSP. It will check and assign
- * the export (to MDT0) for its FID client, so OSP can allocate super sequence
- * or lookup sequence in FLDB of MDT0.
- *
- * \param[in] osp OSP device
- */
-static void osp_prepare_fid_client(struct osp_device *osp)
-{
- LASSERT(osp->opd_obd->u.cli.cl_seq != NULL);
- if (osp->opd_obd->u.cli.cl_seq->lcs_exp != NULL)
- return;
-
- LASSERT(osp->opd_exp != NULL);
- osp->opd_obd->u.cli.cl_seq->lcs_exp =
- class_export_get(osp->opd_exp);
-}
-
-/**
* Implementation of obd_ops::o_import_event
*
* This function is called when some related import event happens. It will
break;
case IMP_EVENT_INACTIVE:
d->opd_imp_active = 0;
+ d->opd_imp_connected = 0;
+ d->opd_obd->obd_inactive = 1;
if (d->opd_connect_mdt)
break;
case IMP_EVENT_ACTIVE:
d->opd_imp_active = 1;
- osp_prepare_fid_client(d);
if (d->opd_got_disconnected)
d->opd_new_connection = 1;
d->opd_imp_connected = 1;
d->opd_imp_seen_connected = 1;
+ d->opd_obd->obd_inactive = 0;
if (d->opd_connect_mdt)
break;
CDEBUG(D_HA, "got connected\n");
break;
case IMP_EVENT_INVALIDATE:
+ if (d->opd_connect_mdt)
+ osp_invalidate_request(d);
+
if (obd->obd_namespace == NULL)
break;
ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
* \retval negative negative errno if the ioctl handling failed.
*/
static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
- void *karg, void *uarg)
+ void *karg, void __user *uarg)
{
struct obd_device *obd = exp->exp_obd;
struct osp_device *d;
return rc;
}
+
/**
* Implementation of obd_ops::o_get_info
*
* \param[in] key the key
* \param[out] vallen length of \a val
* \param[out] val holds the value returned by the key
- * \param[in] unused necessary for the interface but unused
*
* \retval 0 0 if getting information succeeded.
* \retval negative negative errno if getting information failed.
*/
static int osp_obd_get_info(const struct lu_env *env, struct obd_export *exp,
- __u32 keylen, void *key, __u32 *vallen, void *val,
- struct lov_stripe_md *unused)
+ __u32 keylen, void *key, __u32 *vallen, void *val)
{
int rc = -EINVAL;
RETURN(rc);
}
+static int osp_obd_set_info_async(const struct lu_env *env,
+ struct obd_export *exp,
+ u32 keylen, void *key,
+ u32 vallen, void *val,
+ struct ptlrpc_request_set *set)
+{
+ struct obd_device *obd = exp->exp_obd;
+ struct obd_import *imp = obd->u.cli.cl_import;
+ struct osp_device *osp;
+ struct ptlrpc_request *req;
+ char *tmp;
+ int rc;
+
+ if (KEY_IS(KEY_SPTLRPC_CONF)) {
+ sptlrpc_conf_client_adapt(exp->exp_obd);
+ RETURN(0);
+ }
+
+ LASSERT(set != NULL);
+ if (!obd->obd_set_up || obd->obd_stopping)
+ RETURN(-EAGAIN);
+ osp = lu2osp_dev(obd->obd_lu_dev);
+
+ req = ptlrpc_request_alloc(imp, &RQF_OBD_SET_INFO);
+ if (req == NULL)
+ RETURN(-ENOMEM);
+
+ req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
+ RCL_CLIENT, keylen);
+ req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
+ RCL_CLIENT, vallen);
+ if (osp->opd_connect_mdt)
+ rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO);
+ else
+ rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
+ if (rc) {
+ ptlrpc_request_free(req);
+ RETURN(rc);
+ }
+
+ tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
+ memcpy(tmp, key, keylen);
+ tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
+ memcpy(tmp, val, vallen);
+
+ ptlrpc_request_set_replen(req);
+ ptlrpc_set_add_req(set, req);
+ ptlrpc_check_set(NULL, set);
+
+ RETURN(0);
+}
+
/**
* Implementation of obd_ops: o_fid_alloc
*
* requested from seq meta server
* \retval negative negative errno if FID allocation failed.
*/
-int osp_fid_alloc(const struct lu_env *env, struct obd_export *exp,
- struct lu_fid *fid, struct md_op_data *unused)
+static int osp_fid_alloc(const struct lu_env *env, struct obd_export *exp,
+ struct lu_fid *fid, struct md_op_data *unused)
{
struct client_obd *cli = &exp->exp_obd->u.cli;
struct osp_device *osp = lu2osp_dev(exp->exp_obd->obd_lu_dev);
LASSERT(osp->opd_obd->u.cli.cl_seq != NULL);
/* Sigh, fid client is not ready yet */
- if (osp->opd_obd->u.cli.cl_seq->lcs_exp == NULL)
- RETURN(-ENOTCONN);
+ LASSERT(osp->opd_obd->u.cli.cl_seq->lcs_exp != NULL);
RETURN(seq_client_alloc_fid(env, seq, fid));
}
.o_connect = osp_obd_connect,
.o_disconnect = osp_obd_disconnect,
.o_get_info = osp_obd_get_info,
+ .o_set_info_async = osp_obd_set_info_async,
.o_import_event = osp_import_event,
.o_iocontrol = osp_iocontrol,
.o_statfs = osp_obd_statfs,
* \retval 0 0 if initialization succeeds.
* \retval negative negative errno if initialization failed.
*/
-static int __init osp_mod_init(void)
+static int __init osp_init(void)
{
struct obd_type *type;
int rc;
return rc;
type = class_search_type(LUSTRE_OSP_NAME);
- type->typ_procsym = lprocfs_seq_register("osc", proc_lustre_root,
- NULL, NULL);
+ type->typ_procsym = lprocfs_register("osc", proc_lustre_root,
+ NULL, NULL);
if (IS_ERR(type->typ_procsym)) {
CERROR("osp: can't create compat entry \"osc\": %d\n",
(int) PTR_ERR(type->typ_procsym));
* This callback is called when kernel unloads OSP module from memory, and
* it will deregister OSP and LWP device type from obd_types (\see class_obd.c).
*/
-static void __exit osp_mod_exit(void)
+static void __exit osp_exit(void)
{
class_unregister_type(LUSTRE_LWP_NAME);
class_unregister_type(LUSTRE_OSP_NAME);
lu_kmem_fini(osp_caches);
}
-MODULE_AUTHOR("Intel, Inc. <http://www.intel.com/>");
-MODULE_DESCRIPTION("Lustre OST Proxy Device ("LUSTRE_OSP_NAME")");
+MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
+MODULE_DESCRIPTION("Lustre OSD Storage Proxy ("LUSTRE_OSP_NAME")");
+MODULE_VERSION(LUSTRE_VERSION_STRING);
MODULE_LICENSE("GPL");
-cfs_module(osp, LUSTRE_VERSION_STRING, osp_mod_init, osp_mod_exit);
+module_init(osp_init);
+module_exit(osp_exit);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff