* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2010, 2016, Intel Corporation.
+ * Copyright (c) 2010, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_LDLM
+#include <linux/jiffies.h>
#include <linux/kthread.h>
#include <libcfs/libcfs.h>
#include <obd.h>
atomic_long_set(&cli->cl_unstable_count, 0);
INIT_LIST_HEAD(&cli->cl_shrink_list);
+ INIT_LIST_HEAD(&cli->cl_flight_waiters);
+ cli->cl_rpcs_in_flight = 0;
+
init_waitqueue_head(&cli->cl_destroy_waitq);
atomic_set(&cli->cl_destroy_in_flight, 0);
#ifdef ENABLE_CHECKSUM
* from OFD after connecting. */
cli->cl_max_pages_per_rpc = PTLRPC_MAX_BRW_PAGES;
+ cli->cl_short_io_bytes = OBD_MAX_SHORT_IO_BYTES;
+
/* set cl_chunkbits default value to PAGE_SHIFT,
* it will be updated at OSC connection time. */
cli->cl_chunkbits = PAGE_SHIFT;
cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_MAX;
else
cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_DEFAULT;
- }
+ }
spin_lock_init(&cli->cl_mod_rpcs_lock);
spin_lock_init(&cli->cl_mod_rpcs_hist.oh_lock);
ocd->ocd_connect_flags, "old %#llx, new %#llx\n",
data->ocd_connect_flags, ocd->ocd_connect_flags);
data->ocd_connect_flags = ocd->ocd_connect_flags;
+ data->ocd_connect_flags2 = ocd->ocd_connect_flags2;
}
ptlrpc_pinger_add_import(imp);
{
struct obd_device *target;
struct lustre_handle *hdl;
- cfs_time_t now;
- cfs_time_t deadline;
- int timeout;
+ time64_t deadline;
+ time64_t timeout;
+ time64_t now;
int rc = 0;
- ENTRY;
+ ENTRY;
hdl = &exp->exp_imp_reverse->imp_remote_handle;
if (!exp->exp_connection || !lustre_handle_is_used(hdl)) {
conn->cookie = exp->exp_handle.h_cookie;
GOTO(out_already, rc);
}
- now = cfs_time_current();
- deadline = target->obd_recovery_timer.expires;
- if (cfs_time_before(now, deadline)) {
- struct target_distribute_txn_data *tdtd =
- class_exp2tgt(exp)->lut_tdtd;
+ now = ktime_get_seconds();
+ deadline = jiffies_to_msecs(target->obd_recovery_timer.expires) /
+ MSEC_PER_SEC;
+ if (now < deadline) {
+ struct target_distribute_txn_data *tdtd;
int size = 0;
int count = 0;
char *buf = NULL;
- timeout = cfs_duration_sec(cfs_time_sub(deadline, now));
+ timeout = deadline - now;
+ tdtd = class_exp2tgt(exp)->lut_tdtd;
if (tdtd && tdtd->tdtd_show_update_logs_retrievers)
buf = tdtd->tdtd_show_update_logs_retrievers(
tdtd->tdtd_show_retrievers_cbdata,
if (count > 0)
LCONSOLE_WARN("%s: Recovery already passed deadline "
- "%d:%.02d. It is due to DNE recovery "
+ "%lld:%.02lld. It is due to DNE recovery "
"failed/stuck on the %d MDT(s):%s. "
"Please wait until all MDTs recovered "
"or abort the recovery by force.\n",
buf ? buf : "unknown (not enough RAM)");
else
LCONSOLE_WARN("%s: Recovery already passed deadline "
- "%d:%.02d. If you do not want to wait "
+ "%lld:%.02lld. If you do not want to wait "
"more, please abort the recovery by "
"force.\n", target->obd_name,
timeout / 60, timeout % 60);
if (buf != NULL)
OBD_FREE(buf, size);
} else {
- timeout = cfs_duration_sec(cfs_time_sub(now, deadline));
+ timeout = now - deadline;
LCONSOLE_WARN("%s: Recovery already passed deadline"
- " %d:%.02d, It is most likely due to DNE"
+ " %lld:%.02lld, It is most likely due to DNE"
" recovery is failed or stuck, please wait a"
" few more minutes or abort the recovery.\n",
target->obd_name, timeout / 60, timeout % 60);
* reconnect case */
struct lustre_handle conn;
struct lustre_handle *tmp;
- struct obd_uuid tgtuuid;
struct obd_uuid cluuid;
char *str;
int rc = 0;
bool mds_conn = false, lw_client = false, initial_conn = false;
bool mds_mds_conn = false;
bool new_mds_mds_conn = false;
- bool target_referenced = false;
struct obd_connect_data *data, *tmpdata;
int size, tmpsize;
lnet_nid_t *client_nid = NULL;
GOTO(out, rc = -EINVAL);
}
- obd_str2uuid(&tgtuuid, str);
- target = class_uuid2obd(&tgtuuid);
- if (!target)
- target = class_name2obd(str);
-
+ target = class_dev_by_str(str);
if (!target) {
deuuidify(str, NULL, &target_start, &target_len);
LCONSOLE_ERROR_MSG(0x137, "%s: not available for connect "
}
spin_lock(&target->obd_dev_lock);
+
+ target->obd_conn_inprogress++;
+
if (target->obd_stopping || !target->obd_set_up) {
spin_unlock(&target->obd_dev_lock);
GOTO(out, rc = -EAGAIN);
}
- /* Make sure the target isn't cleaned up while we're here. Yes,
- * there's still a race between the above check and our incref here.
- * Really, class_uuid2obd should take the ref. */
- class_incref(target, __func__, current);
- target_referenced = true;
-
- target->obd_conn_inprogress++;
spin_unlock(&target->obd_dev_lock);
str = req_capsule_client_get(&req->rq_pill, &RMF_CLUUID);
*/
if (!(data->ocd_connect_flags & OBD_CONNECT_FULL20))
GOTO(out, rc = -EPROTO);
-#endif
+ /* Don't allow liblustre clients to connect.
+ * - testing was disabled in v2_2_50_0-61-g6a75d65
+ * - building was disabled in v2_5_58_0-28-g7277179
+ * - client code was deleted in v2_6_50_0-101-gcdfbc72,
+ * - clients were refused connect for version difference > 0.0.1.32 */
if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
- if (data->ocd_version < LUSTRE_VERSION_CODE -
- LUSTRE_VERSION_ALLOWED_OFFSET ||
- data->ocd_version > LUSTRE_VERSION_CODE +
- LUSTRE_VERSION_ALLOWED_OFFSET) {
- DEBUG_REQ(D_WARNING, req, "Refusing %s (%d.%d.%d.%d) "
- "libclient connection attempt",
- data->ocd_version < LUSTRE_VERSION_CODE ?
- "old" : "new",
- OBD_OCD_VERSION_MAJOR(data->ocd_version),
- OBD_OCD_VERSION_MINOR(data->ocd_version),
- OBD_OCD_VERSION_PATCH(data->ocd_version),
- OBD_OCD_VERSION_FIX(data->ocd_version));
- data = req_capsule_server_sized_get(&req->rq_pill,
- &RMF_CONNECT_DATA,
- offsetof(typeof(*data), ocd_version) +
- sizeof(data->ocd_version));
- if (data) {
- data->ocd_connect_flags = OBD_CONNECT_VERSION;
- data->ocd_version = LUSTRE_VERSION_CODE;
- }
- GOTO(out, rc = -EPROTO);
- }
+ DEBUG_REQ(D_WARNING, req, "Refusing libclient connection");
+ GOTO(out, rc = -EPROTO);
}
+#endif
/* Note: lw_client is needed in MDS-MDS failover during update log
* processing, so we needs to allow lw_client to be connected at
*
* Via check OBD_CONNECT_FID, we can distinguish whether
* the OBD_CONNECT_MDS_MDS/OBD_CONNECT_MNE_SWAB is from
- * MGC or MDT. */
+ * MGC or MDT, since MGC does not use OBD_CONNECT_FID.
+ */
if (!lw_client &&
(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
(data->ocd_connect_flags & OBD_CONNECT_FID) &&
GOTO(out, rc);
}
- CDEBUG(D_HA, "%s: connection from %s@%s %st%llu exp %p cur %ld last %ld\n",
- target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
- target->obd_recovering ? "recovering/" : "", data->ocd_transno,
- export, (long)cfs_time_current_sec(),
- export ? (long)export->exp_last_request_time : 0);
+ CDEBUG(D_HA, "%s: connection from %s@%s %st%llu exp %p cur %lld last %lld\n",
+ target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
+ target->obd_recovering ? "recovering/" : "", data->ocd_transno,
+ export, ktime_get_real_seconds(),
+ export ? export->exp_last_request_time : 0);
/* If this is the first time a client connects, reset the recovery
* timer. Discard lightweight connections which might be local. */
/* allow "new" MDT to be connected during recovery, since we
* need retrieve recovery update records from it */
if (target->obd_recovering && !lw_client && !mds_mds_conn) {
- cfs_time_t t;
- int c; /* connected */
- int i; /* in progress */
- int k; /* known */
- int s; /* stale/evicted */
+ time64_t t;
+ int c; /* connected */
+ int i; /* in progress */
+ int k; /* known */
+ int s; /* stale/evicted */
c = atomic_read(&target->obd_connected_clients);
i = atomic_read(&target->obd_lock_replay_clients);
k = target->obd_max_recoverable_clients;
s = target->obd_stale_clients;
- t = target->obd_recovery_timer.expires;
- t = cfs_time_sub(t, cfs_time_current());
- t = cfs_duration_sec(t);
+ t = jiffies_to_msecs(target->obd_recovery_timer.expires);
+ t /= MSEC_PER_SEC;
+ t -= ktime_get_seconds();
LCONSOLE_WARN("%s: Denying connection for new client %s"
"(at %s), waiting for %d known clients "
"(%d recovered, %d in progress, and %d "
- "evicted) to recover in %d:%.02d\n",
+ "evicted) to recover in %lld:%.02lld\n",
target->obd_name, cluuid.uuid,
libcfs_nid2str(req->rq_peer.nid), k,
- c - i, i, s, (int)t / 60,
- (int)t % 60);
+ c - i, i, s, t / 60, t % 60);
rc = -EBUSY;
} else {
dont_check_exports:
spin_unlock(&export->exp_lock);
CDEBUG(D_RPCTRACE, "%s: %s already connected at greater "
"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));
-
- GOTO(out, rc = -EALREADY);
- }
- LASSERT(lustre_msg_get_conn_cnt(req->rq_reqmsg) > 0);
- export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
+ cluuid.uuid, libcfs_nid2str(req->rq_peer.nid),
+ export->exp_conn_cnt,
+ lustre_msg_get_conn_cnt(req->rq_reqmsg));
- /* Don't evict liblustre clients for not pinging. */
- if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
- export->exp_libclient = 1;
- spin_unlock(&export->exp_lock);
-
- spin_lock(&target->obd_dev_lock);
- list_del_init(&export->exp_obd_chain_timed);
- spin_unlock(&target->obd_dev_lock);
- } else {
- spin_unlock(&export->exp_lock);
+ GOTO(out, rc = -EALREADY);
}
+ LASSERT(lustre_msg_get_conn_cnt(req->rq_reqmsg) > 0);
+ export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
+ spin_unlock(&export->exp_lock);
- if (export->exp_connection != NULL) {
+ if (export->exp_connection != NULL) {
/* Check to see if connection came from another NID. */
- if ((export->exp_connection->c_peer.nid != req->rq_peer.nid) &&
+ if ((export->exp_connection->c_peer.nid != req->rq_peer.nid) &&
!hlist_unhashed(&export->exp_nid_hash))
- cfs_hash_del(export->exp_obd->obd_nid_hash,
- &export->exp_connection->c_peer.nid,
- &export->exp_nid_hash);
+ cfs_hash_del(export->exp_obd->obd_nid_hash,
+ &export->exp_connection->c_peer.nid,
+ &export->exp_nid_hash);
- ptlrpc_connection_put(export->exp_connection);
- }
+ ptlrpc_connection_put(export->exp_connection);
+ }
export->exp_connection = ptlrpc_connection_get(req->rq_peer,
req->rq_self,
class_export_put(export);
}
- if (target_referenced == true && target != NULL) {
+ if (target != NULL) {
spin_lock(&target->obd_dev_lock);
target->obd_conn_inprogress--;
spin_unlock(&target->obd_dev_lock);
-
- class_decref(target, __func__, current);
+ class_decref(target, "find", current);
}
req->rq_status = rc;
RETURN(rc);
obd->obd_recovery_end = ktime_get_real_seconds();
/* When recovery finished, cleanup orphans on MDS and OST. */
- if (OBT(obd) && OBP(obd, postrecov)) {
- int rc = OBP(obd, postrecov)(obd);
- if (rc < 0)
- LCONSOLE_WARN("%s: Post recovery failed, rc %d\n",
- obd->obd_name, rc);
- }
+ if (obd->obd_type && OBP(obd, postrecov)) {
+ int rc = OBP(obd, postrecov)(obd);
+
+ if (rc < 0)
+ LCONSOLE_WARN("%s: Post recovery failed, rc %d\n",
+ obd->obd_name, rc);
+ }
EXIT;
}
}
mod_timer(&obd->obd_recovery_timer,
- cfs_time_shift(obd->obd_recovery_timeout));
+ jiffies + cfs_time_seconds(obd->obd_recovery_timeout));
obd->obd_recovery_start = ktime_get_real_seconds();
spin_unlock(&obd->obd_dev_lock);
* if @extend is true, extend recovery window to have @drt remaining at least;
* otherwise, make sure the recovery timeout value is not less than @drt.
*/
-static void extend_recovery_timer(struct obd_device *obd, int drt,
+static void extend_recovery_timer(struct obd_device *obd, time64_t drt,
bool extend)
{
time64_t now;
obd->obd_recovery_timeout = to;
end = obd->obd_recovery_start + to;
mod_timer(&obd->obd_recovery_timer,
- cfs_time_shift(end - now));
+ jiffies + cfs_time_seconds(end - now));
}
spin_unlock(&obd->obd_dev_lock);
struct ptlrpc_request *req,
int new_client)
{
- int service_time = lustre_msg_get_service_time(req->rq_reqmsg);
+ time64_t service_time = lustre_msg_get_service_time(req->rq_reqmsg);
struct obd_device_target *obt = &obd->u.obt;
if (!new_client && service_time)
target_start_recovery_timer(obd);
/* Convert the service time to RPC timeout,
- * and reuse service_time to limit stack usage. */
+ * and reuse service_time to limit stack usage.
+ */
service_time = at_est2timeout(service_time);
if (OBD_FAIL_CHECK(OBD_FAIL_TGT_SLUGGISH_NET) &&
{
struct obd_device *obd = lut->lut_obd;
struct target_distribute_txn_data *tdtd;
+ time64_t last = 0;
+ time64_t now;
repeat:
+ if (obd->obd_recovering && obd->obd_recovery_start == 0) {
+ now = ktime_get_seconds();
+ if (now - last > 600) {
+ LCONSOLE_INFO("%s: in recovery but waiting for "
+ "the first client to connect\n",
+ obd->obd_name);
+ last = now;
+ }
+ }
if (obd->obd_recovery_start != 0 && ktime_get_real_seconds() >=
(obd->obd_recovery_start + obd->obd_recovery_time_hard)) {
__u64 next_update_transno = 0;
/* don't reset timer for final stage */
if (!exp_finished(req->rq_export)) {
- int to = obd_timeout;
+ time64_t to = obd_timeout;
/**
* Add request timeout to the recovery time so next request from
target_process_req_flags(obd, req);
if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) {
+ if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_TGT_RECOVERY_REQ_RACE))) {
+ if (cfs_fail_val == 1) {
+ cfs_race_state = 1;
+ cfs_fail_val = 0;
+ wake_up(&cfs_race_waitq);
+
+ set_current_state(TASK_INTERRUPTIBLE);
+ schedule_timeout(cfs_time_seconds(1));
+ }
+ }
+
/* client declares he's ready to complete recovery
* so, we put the request on th final queue */
target_request_copy_get(req);
int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc,
struct l_wait_info *lwi)
{
- struct ptlrpc_request *req = desc->bd_req;
- time_t start = cfs_time_current_sec();
- time_t deadline;
- int rc = 0;
+ struct ptlrpc_request *req = desc->bd_req;
+ time64_t start = ktime_get_real_seconds();
+ time64_t deadline;
+ int rc = 0;
ENTRY;
deadline = req->rq_deadline;
do {
- long timeoutl = deadline - cfs_time_current_sec();
- cfs_duration_t timeout = timeoutl <= 0 ?
- CFS_TICK : cfs_time_seconds(timeoutl);
- time_t rq_deadline;
+ time64_t timeoutl = deadline - ktime_get_real_seconds();
+ long timeout_jiffies = timeoutl <= 0 ?
+ 1 : cfs_time_seconds(timeoutl);
+ time64_t rq_deadline;
- *lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
+ *lwi = LWI_TIMEOUT_INTERVAL(timeout_jiffies,
+ cfs_time_seconds(1),
target_bulk_timeout, desc);
rc = l_wait_event(desc->bd_waitq,
!ptlrpc_server_bulk_active(desc) ||
lwi);
LASSERT(rc == 0 || rc == -ETIMEDOUT);
/* Wait again if we changed rq_deadline. */
- rq_deadline = ACCESS_ONCE(req->rq_deadline);
+ rq_deadline = READ_ONCE(req->rq_deadline);
deadline = start + bulk_timeout;
if (deadline > rq_deadline)
deadline = rq_deadline;
- } while ((rc == -ETIMEDOUT) &&
- (deadline > cfs_time_current_sec()));
+ } while (rc == -ETIMEDOUT &&
+ deadline > ktime_get_real_seconds());
if (rc == -ETIMEDOUT) {
- DEBUG_REQ(D_ERROR, req, "timeout on bulk %s after %ld%+lds",
+ DEBUG_REQ(D_ERROR, req, "timeout on bulk %s after %lld%+llds",
bulk2type(req), deadline - start,
- cfs_time_current_sec() - deadline);
+ ktime_get_real_seconds() - deadline);
ptlrpc_abort_bulk(desc);
} else if (exp->exp_failed) {
DEBUG_REQ(D_ERROR, req, "Eviction on bulk %s",