Whamcloud - gitweb
git://git.whamcloud.com
/
fs
/
lustre-release.git
/ blobdiff
commit
grep
author
committer
pickaxe
?
search:
re
summary
|
shortlog
|
log
|
commit
|
commitdiff
|
tree
raw
| inline |
side by side
LU-1944 ldlm: Add another net_latency to recoverry timer.
[fs/lustre-release.git]
/
lustre
/
ldlm
/
ldlm_lib.c
diff --git
a/lustre/ldlm/ldlm_lib.c
b/lustre/ldlm/ldlm_lib.c
index
2fa220b
..
a7ada1b
100644
(file)
--- a/
lustre/ldlm/ldlm_lib.c
+++ b/
lustre/ldlm/ldlm_lib.c
@@
-131,6
+131,7
@@
int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
{
return import_set_conn(imp, uuid, priority, 1);
}
+EXPORT_SYMBOL(client_import_add_conn);
int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid)
{
@@
-186,6
+187,7
@@
out:
CERROR("connection %s not found\n", uuid->uuid);
RETURN(rc);
}
+EXPORT_SYMBOL(client_import_del_conn);
/**
* Find conn uuid by peer nid. @peer is a server nid. This function is used
@@
-266,7
+268,9
@@
int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
/* In a more perfect world, we would hang a ptlrpc_client off of
* obd_type and just use the values from there. */
- if (!strcmp(name, LUSTRE_OSC_NAME)) {
+ if (!strcmp(name, LUSTRE_OSC_NAME) ||
+ (!strcmp(name, LUSTRE_OSP_NAME) &&
+ !is_osp_on_ost(lustre_cfg_buf(lcfg, 0)))) {
rq_portal = OST_REQUEST_PORTAL;
rp_portal = OSC_REPLY_PORTAL;
connect_op = OST_CONNECT;
@@
-274,7
+278,9
@@
int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
cli->cl_sp_to = LUSTRE_SP_OST;
ns_type = LDLM_NS_TYPE_OSC;
- } else if (!strcmp(name, LUSTRE_MDC_NAME)) {
+ } else if (!strcmp(name, LUSTRE_MDC_NAME) ||
+ (!strcmp(name, LUSTRE_OSP_NAME) &&
+ is_osp_on_ost(lustre_cfg_buf(lcfg, 0)))) {
rq_portal = MDS_REQUEST_PORTAL;
rp_portal = MDC_REPLY_PORTAL;
connect_op = MDS_CONNECT;
@@
-291,6
+297,14
@@
int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
cli->cl_flvr_mgc.sf_rpc = SPTLRPC_FLVR_INVALID;
ns_type = LDLM_NS_TYPE_MGC;
+ } else if (!strcmp(name, LUSTRE_OSP_NAME)) {
+ rq_portal = OST_REQUEST_PORTAL;
+ rp_portal = OSC_REPLY_PORTAL;
+ connect_op = OST_CONNECT;
+ cli->cl_sp_me = LUSTRE_SP_CLI;
+ cli->cl_sp_to = LUSTRE_SP_OST;
+ ns_type = LDLM_NS_TYPE_OSC;
+
} else {
CERROR("unknown client OBD type \"%s\", can't setup\n",
name);
@@
-347,6
+361,15
@@
int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
cfs_spin_lock_init(&cli->cl_write_page_hist.oh_lock);
cfs_spin_lock_init(&cli->cl_read_offset_hist.oh_lock);
cfs_spin_lock_init(&cli->cl_write_offset_hist.oh_lock);
+
+ /* lru for osc. */
+ CFS_INIT_LIST_HEAD(&cli->cl_lru_osc);
+ cfs_atomic_set(&cli->cl_lru_shrinkers, 0);
+ cfs_atomic_set(&cli->cl_lru_busy, 0);
+ cfs_atomic_set(&cli->cl_lru_in_list, 0);
+ CFS_INIT_LIST_HEAD(&cli->cl_lru_list);
+ client_obd_list_lock_init(&cli->cl_lru_list_lock);
+
cfs_waitq_init(&cli->cl_destroy_waitq);
cfs_atomic_set(&cli->cl_destroy_in_flight, 0);
#ifdef ENABLE_CHECKSUM
@@
-443,6
+466,7
@@
err:
RETURN(rc);
}
+EXPORT_SYMBOL(client_obd_setup);
int client_obd_cleanup(struct obd_device *obddev)
{
@@
-456,6
+480,7
@@
int client_obd_cleanup(struct obd_device *obddev)
ldlm_put_ref();
RETURN(0);
}
+EXPORT_SYMBOL(client_obd_cleanup);
/* ->o_connect() method for client side (OSC and MDC and MGC) */
int client_connect_import(const struct lu_env *env,
@@
-524,6
+549,7
@@
out_sem:
return rc;
}
+EXPORT_SYMBOL(client_connect_import);
int client_disconnect_export(struct obd_export *exp)
{
@@
-598,6
+624,7
@@
int client_disconnect_export(struct obd_export *exp)
RETURN(rc);
}
+EXPORT_SYMBOL(client_disconnect_export);
#ifdef HAVE_SERVER_SUPPORT
int server_disconnect_export(struct obd_export *exp)
@@
-635,6
+662,7
@@
int server_disconnect_export(struct obd_export *exp)
RETURN(rc);
}
+EXPORT_SYMBOL(server_disconnect_export);
/* --------------------------------------------------------------------------
* from old lib/target.c
@@
-744,7
+772,7
@@
int target_handle_connect(struct ptlrpc_request *req)
int rc = 0;
char *target_start;
int target_len;
-
int mds_conn = 0
;
+
bool mds_conn = false, lw_client = false
;
struct obd_connect_data *data, *tmpdata;
int size, tmpsize;
lnet_nid_t *client_nid = NULL;
@@
-870,7
+898,10
@@
int target_handle_connect(struct ptlrpc_request *req)
if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL) &&
(data->ocd_connect_flags & OBD_CONNECT_MDS))
- mds_conn = 1;
+ mds_conn = true;
+
+ if ((data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) != 0)
+ lw_client = true;
/* lctl gets a backstage, all-access pass. */
if (obd_uuid_equals(&cluuid, &target->obd_uuid))
@@
-881,7
+912,11
@@
int target_handle_connect(struct ptlrpc_request *req)
goto no_export;
/* we've found an export in the hash */
+
+ cfs_spin_lock(&export->exp_lock);
+
if (export->exp_connecting) { /* bug 9635, et. al. */
+ cfs_spin_unlock(&export->exp_lock);
LCONSOLE_WARN("%s: Export %p already connecting from %s\n",
export->exp_obd->obd_name, export,
libcfs_nid2str(req->rq_peer.nid));
@@
-889,6
+924,7
@@
int target_handle_connect(struct ptlrpc_request *req)
export = NULL;
rc = -EALREADY;
} else if (mds_conn && export->exp_connection) {
+ cfs_spin_unlock(&export->exp_lock);
if (req->rq_peer.nid != export->exp_connection->c_peer.nid)
/* mds reconnected after failover */
LCONSOLE_WARN("%s: Received MDS connection from "
@@
-908,6
+944,7
@@
int target_handle_connect(struct ptlrpc_request *req)
req->rq_peer.nid != export->exp_connection->c_peer.nid &&
(lustre_msg_get_op_flags(req->rq_reqmsg) &
MSG_CONNECT_INITIAL)) {
+ cfs_spin_unlock(&export->exp_lock);
/* in mds failover we have static uuid but nid can be
* changed*/
LCONSOLE_WARN("%s: Client %s seen on new nid %s when "
@@
-920,13
+957,12
@@
int target_handle_connect(struct ptlrpc_request *req)
class_export_put(export);
export = NULL;
} else {
- cfs_spin_lock(&export->exp_lock);
- export->exp_connecting = 1;
- cfs_spin_unlock(&export->exp_lock);
- LASSERT(export->exp_obd == target);
+ export->exp_connecting = 1;
+ cfs_spin_unlock(&export->exp_lock);
+ LASSERT(export->exp_obd == target);
-
rc = target_handle_reconnect(&conn, export, &cluuid);
-
}
+ rc = target_handle_reconnect(&conn, export, &cluuid);
+ }
/* If we found an export, we already unlocked. */
if (!export) {
@@
-978,8
+1014,8
@@
no_export:
export ? (long)export->exp_last_request_time : 0);
/* If this is the first time a client connects, reset the recovery
-
* timer
*/
-
if (
rc == 0 && target->obd_recovering)
+
* timer. Discard lightweight connections which might be local
*/
+
if (!lw_client &&
rc == 0 && target->obd_recovering)
check_and_start_recovery_timer(target, req, export == NULL);
/* We want to handle EALREADY but *not* -EALREADY from
@@
-997,7
+1033,8
@@
no_export:
client_nid = &req->rq_peer.nid;
if (export == NULL) {
- if (target->obd_recovering) {
+ /* allow lightweight connections during recovery */
+ if (target->obd_recovering && !lw_client) {
cfs_time_t t;
int c; /* connected */
int i; /* in progress */
@@
-1084,8
+1121,8
@@
dont_check_exports:
cfs_spin_lock(&export->exp_lock);
if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
cfs_spin_unlock(&export->exp_lock);
-
CDEBUG(D_RPCTRACE, "%s: %s already connected at high
er "
-
"conn_cnt: %d >
%d\n",
+
CDEBUG(D_RPCTRACE, "%s: %s already connected at great
er "
+
"or equal conn_cnt: %d >=
%d\n",
cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
export->exp_conn_cnt,
lustre_msg_get_conn_cnt(req->rq_reqmsg));
@@
-1128,7
+1165,7
@@
dont_check_exports:
&export->exp_nid_hash);
}
-
if (target->obd_recovering && !export->exp_in_recovery
) {
+
if (target->obd_recovering && !export->exp_in_recovery && !lw_client
) {
int has_transno;
__u64 transno = data->ocd_transno;
@@
-1168,7
+1205,7
@@
dont_check_exports:
}
/* Tell the client we're in recovery, when client is involved in it. */
-
if (target->obd_recovering
)
+
if (target->obd_recovering && !lw_client
)
lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN);
@@
-1184,13
+1221,9
@@
dont_check_exports:
}
cfs_spin_lock(&export->exp_lock);
- if (export->exp_imp_reverse != NULL)
{
+ if (export->exp_imp_reverse != NULL)
/* destroyed import can be still referenced in ctxt */
- obd_set_info_async(req->rq_svc_thread->t_env, export,
- sizeof(KEY_REVIMP_UPD), KEY_REVIMP_UPD,
- 0, NULL, NULL);
tmp_imp = export->exp_imp_reverse;
- }
export->exp_imp_reverse = revimp;
cfs_spin_unlock(&export->exp_lock);
@@
-1204,13
+1237,13
@@
dont_check_exports:
* ptlrpc_handle_server_req_in->lustre_unpack_msg() */
revimp->imp_msg_magic = req->rq_reqmsg->lm_magic;
-
if ((export->exp
_connect_flags & OBD_CONNECT_AT) &&
-
(revimp->imp_msg_magic != LUSTRE_MSG_MAGIC_V1))
-
revimp->imp_msghdr_flags |= MSGHDR_AT_SUPPORT;
-
else
-
revimp->imp_msghdr_flags &= ~MSGHDR_AT_SUPPORT;
+
if ((data->ocd
_connect_flags & OBD_CONNECT_AT) &&
+ (revimp->imp_msg_magic != LUSTRE_MSG_MAGIC_V1))
+ revimp->imp_msghdr_flags |= MSGHDR_AT_SUPPORT;
+ else
+ revimp->imp_msghdr_flags &= ~MSGHDR_AT_SUPPORT;
-
if ((export->exp
_connect_flags & OBD_CONNECT_FULL20) &&
+
if ((data->ocd
_connect_flags & OBD_CONNECT_FULL20) &&
(revimp->imp_msg_magic != LUSTRE_MSG_MAGIC_V1))
revimp->imp_msghdr_flags |= MSGHDR_CKSUM_INCOMPAT18;
else
@@
-1248,6
+1281,7
@@
out:
req->rq_status = rc;
RETURN(rc);
}
+EXPORT_SYMBOL(target_handle_connect);
int target_handle_disconnect(struct ptlrpc_request *req)
{
@@
-1263,6
+1297,7
@@
int target_handle_disconnect(struct ptlrpc_request *req)
RETURN(0);
}
+EXPORT_SYMBOL(target_handle_disconnect);
void target_destroy_export(struct obd_export *exp)
{
@@
-1283,6
+1318,7
@@
void target_destroy_export(struct obd_export *exp)
LASSERT_ATOMIC_ZERO(&exp->exp_cb_count);
LASSERT_ATOMIC_ZERO(&exp->exp_replay_count);
}
+EXPORT_SYMBOL(target_destroy_export);
/*
* Recovery functions
@@
-1355,17
+1391,20
@@
static void target_exp_dequeue_req_replay(struct ptlrpc_request *req)
#ifdef __KERNEL__
static void target_finish_recovery(struct obd_device *obd)
{
- time_t elapsed_time = max_t(time_t, 1, cfs_time_current_sec() -
- obd->obd_recovery_start);
ENTRY;
- LCONSOLE_INFO("%s: Recovery over after %d:%.02d, of %d clients "
- "%d recovered and %d %s evicted.\n", obd->obd_name,
- (int)elapsed_time / 60, (int)elapsed_time % 60,
- obd->obd_max_recoverable_clients,
- cfs_atomic_read(&obd->obd_connected_clients),
- obd->obd_stale_clients,
- obd->obd_stale_clients == 1 ? "was" : "were");
+ /* only log a recovery message when recovery has occurred */
+ if (obd->obd_recovery_start) {
+ time_t elapsed_time = max_t(time_t, 1, cfs_time_current_sec() -
+ obd->obd_recovery_start);
+ LCONSOLE_INFO("%s: Recovery over after %d:%.02d, of %d clients "
+ "%d recovered and %d %s evicted.\n", obd->obd_name,
+ (int)elapsed_time / 60, (int)elapsed_time % 60,
+ obd->obd_max_recoverable_clients,
+ cfs_atomic_read(&obd->obd_connected_clients),
+ obd->obd_stale_clients,
+ obd->obd_stale_clients == 1 ? "was" : "were");
+ }
ldlm_reprocess_all_ns(obd->obd_namespace);
cfs_spin_lock(&obd->obd_recovery_task_lock);
@@
-1485,6
+1524,7
@@
void target_cleanup_recovery(struct obd_device *obd)
EXIT;
}
+EXPORT_SYMBOL(target_cleanup_recovery);
/* obd_recovery_task_lock should be held */
void target_cancel_recovery_timer(struct obd_device *obd)
@@
-1492,6
+1532,7
@@
void target_cancel_recovery_timer(struct obd_device *obd)
CDEBUG(D_HA, "%s: cancel recovery timer\n", obd->obd_name);
cfs_timer_disarm(&obd->obd_recovery_timer);
}
+EXPORT_SYMBOL(target_cancel_recovery_timer);
static void target_start_recovery_timer(struct obd_device *obd)
{
@@
-1917,8
+1958,11
@@
static int handle_recovery_req(struct ptlrpc_thread *thread,
to = max((int)at_est2timeout(
at_get(&svcpt->scp_at_estimate)),
(int)lustre_msg_get_timeout(req->rq_reqmsg));
- /* Add net_latency (see ptlrpc_replay_req) */
- to += lustre_msg_get_service_time(req->rq_reqmsg);
+ /* Add 2 net_latency, one for balance rq_deadline
+ * (see ptl_send_rpc), one for resend the req to server,
+ * Note: client will pack net_latency in replay req
+ * (see ptlrpc_replay_req) */
+ to += 2 * lustre_msg_get_service_time(req->rq_reqmsg);
}
extend_recovery_timer(class_exp2obd(req->rq_export), to, true);
}
@@
-2030,7
+2074,7
@@
static int target_recovery_thread(void *arg)
*/
CDEBUG(D_INFO, "3: final stage - process recovery completion pings\n");
/** Update server last boot epoch */
-
lu
t_boot_epoch_update(lut);
+
tg
t_boot_epoch_update(lut);
/* We drop recoverying flag to forward all new requests
* to regular mds_handle() since now */
cfs_spin_lock(&obd->obd_dev_lock);
@@
-2103,6
+2147,7
@@
void target_stop_recovery_thread(struct obd_device *obd)
cfs_wait_for_completion(&trd->trd_finishing);
}
}
+EXPORT_SYMBOL(target_stop_recovery_thread);
void target_recovery_fini(struct obd_device *obd)
{
@@
-2130,7
+2175,7
@@
void target_recovery_init(struct lu_target *lut, svc_handler_t handler)
struct obd_device *obd = lut->lut_obd;
if (obd->obd_max_recoverable_clients == 0) {
/** Update server last boot epoch */
-
lu
t_boot_epoch_update(lut);
+
tg
t_boot_epoch_update(lut);
return;
}
@@
-2316,12
+2361,14
@@
int target_queue_recovery_request(struct ptlrpc_request *req,
cfs_waitq_signal(&obd->obd_next_transno_waitq);
RETURN(0);
}
+EXPORT_SYMBOL(target_queue_recovery_request);
int target_handle_ping(struct ptlrpc_request *req)
{
obd_ping(req->rq_svc_thread->t_env, req->rq_export);
return req_capsule_server_pack(&req->rq_pill);
}
+EXPORT_SYMBOL(target_handle_ping);
void target_committed_to_req(struct ptlrpc_request *req)
{
@@
-2373,6
+2420,7
@@
int target_pack_pool_reply(struct ptlrpc_request *req)
RETURN(0);
}
+EXPORT_SYMBOL(target_pack_pool_reply);
int target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id)
{
@@
-2436,7
+2484,7
@@
void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id)
rs->rs_xid = req->rq_xid;
rs->rs_transno = req->rq_transno;
rs->rs_export = exp;
- rs->rs_opc = lustre_msg_get_opc(r
s->rs_
msg);
+ rs->rs_opc = lustre_msg_get_opc(r
eq->rq_req
msg);
cfs_spin_lock(&exp->exp_uncommitted_replies_lock);
CDEBUG(D_NET, "rs transno = "LPU64", last committed = "LPU64"\n",
@@
-2483,119
+2531,7
@@
void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id)
cfs_spin_unlock(&svcpt->scp_rep_lock);
EXIT;
}
-
-int target_handle_qc_callback(struct ptlrpc_request *req)
-{
- struct obd_quotactl *oqctl;
- struct client_obd *cli = &req->rq_export->exp_obd->u.cli;
-
- oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
- if (oqctl == NULL) {
- CERROR("Can't unpack obd_quotactl\n");
- RETURN(-EPROTO);
- }
-
- cli->cl_qchk_stat = oqctl->qc_stat;
-
- return 0;
-}
-
-#ifdef HAVE_QUOTA_SUPPORT
-int target_handle_dqacq_callback(struct ptlrpc_request *req)
-{
-#ifdef __KERNEL__
- struct obd_device *obd = req->rq_export->exp_obd;
- struct obd_device *master_obd = NULL, *lov_obd = NULL;
- struct obd_device_target *obt;
- struct lustre_quota_ctxt *qctxt;
- struct qunit_data *qdata = NULL;
- int rc = 0;
- ENTRY;
-
- if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DROP_QUOTA_REQ))
- RETURN(rc);
-
- rc = req_capsule_server_pack(&req->rq_pill);
- if (rc) {
- CERROR("packing reply failed!: rc = %d\n", rc);
- RETURN(rc);
- }
-
- LASSERT(req->rq_export);
-
- qdata = quota_get_qdata(req, QUOTA_REQUEST, QUOTA_EXPORT);
- if (IS_ERR(qdata)) {
- rc = PTR_ERR(qdata);
- CDEBUG(D_ERROR, "Can't unpack qunit_data(rc: %d)\n", rc);
- req->rq_status = rc;
- GOTO(out, rc);
- }
-
- /* we use the observer */
- if (obd_pin_observer(obd, &lov_obd) ||
- obd_pin_observer(lov_obd, &master_obd)) {
- CERROR("Can't find the observer, it is recovering\n");
- req->rq_status = -EAGAIN;
- GOTO(out, rc);
- }
-
- obt = &master_obd->u.obt;
- qctxt = &obt->obt_qctxt;
-
- if (!qctxt->lqc_setup || !qctxt->lqc_valid) {
- /* quota_type has not been processed yet, return EAGAIN
- * until we know whether or not quotas are supposed to
- * be enabled */
- CDEBUG(D_QUOTA, "quota_type not processed yet, return "
- "-EAGAIN\n");
- req->rq_status = -EAGAIN;
- GOTO(out, rc);
- }
-
- cfs_down_read(&obt->obt_rwsem);
- if (qctxt->lqc_lqs_hash == NULL) {
- cfs_up_read(&obt->obt_rwsem);
- /* quota_type has not been processed yet, return EAGAIN
- * until we know whether or not quotas are supposed to
- * be enabled */
- CDEBUG(D_QUOTA, "quota_ctxt is not ready yet, return "
- "-EAGAIN\n");
- req->rq_status = -EAGAIN;
- GOTO(out, rc);
- }
-
- LASSERT(qctxt->lqc_handler);
- rc = qctxt->lqc_handler(master_obd, qdata,
- lustre_msg_get_opc(req->rq_reqmsg));
- cfs_up_read(&obt->obt_rwsem);
- if (rc && rc != -EDQUOT)
- CDEBUG(rc == -EBUSY ? D_QUOTA : D_ERROR,
- "dqacq/dqrel failed! (rc:%d)\n", rc);
- req->rq_status = rc;
-
- rc = quota_copy_qdata(req, qdata, QUOTA_REPLY, QUOTA_EXPORT);
- if (rc < 0) {
- CERROR("Can't pack qunit_data(rc: %d)\n", rc);
- GOTO(out, rc);
- }
-
- /* Block the quota req. b=14840 */
- OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_BLOCK_QUOTA_REQ, obd_timeout);
- EXIT;
-
-out:
- if (master_obd)
- obd_unpin_observer(lov_obd);
- if (lov_obd)
- obd_unpin_observer(obd);
-
- rc = ptlrpc_reply(req);
- return rc;
-#else
- return 0;
-#endif /* !__KERNEL__ */
-}
-#endif /* HAVE_QUOTA_SUPPORT */
+EXPORT_SYMBOL(target_send_reply);
ldlm_mode_t lck_compat_array[] = {
[LCK_EX] LCK_COMPAT_EX,