sizeof(server_uuid)));
cli->cl_dirty_pages = 0;
+ cli->cl_dirty_max_pages = 0;
cli->cl_avail_grant = 0;
/* FIXME: Should limit this for the sum of all cl_dirty_max_pages. */
/* cl_dirty_max_pages may be changed at connect time in
spin_lock_init(&cli->cl_lru_list_lock);
atomic_long_set(&cli->cl_unstable_count, 0);
INIT_LIST_HEAD(&cli->cl_shrink_list);
+ INIT_LIST_HEAD(&cli->cl_grant_chain);
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);
+
+ cli->cl_supp_cksum_types = OBD_CKSUM_CRC32;
#ifdef ENABLE_CHECKSUM
/* Turn on checksumming by default. */
cli->cl_checksum = 1;
* Set cl_chksum* to CRC32 for now to avoid returning screwed info
* through procfs.
*/
- cli->cl_cksum_type = cli->cl_supp_cksum_types = OBD_CKSUM_CRC32;
+ cli->cl_cksum_type = cli->cl_supp_cksum_types;
#endif
atomic_set(&cli->cl_resends, OSC_DEFAULT_RESENDS);
* from OFD after connecting. */
cli->cl_max_pages_per_rpc = PTLRPC_MAX_BRW_PAGES;
- cli->cl_short_io_bytes = OBD_MAX_SHORT_IO_BYTES;
+ cli->cl_max_short_io_bytes = OBD_MAX_SHORT_IO_BYTES;
/* set cl_chunkbits default value to PAGE_SHIFT,
* it will be updated at OSC connection time. */
{
struct obd_device *target;
struct lustre_handle *hdl;
- time64_t deadline;
- time64_t timeout;
- time64_t now;
+ ktime_t remaining;
+ s64 timeout;
int rc = 0;
ENTRY;
GOTO(out_already, rc);
}
- now = ktime_get_seconds();
- deadline = jiffies_to_msecs(target->obd_recovery_timer.expires) /
- MSEC_PER_SEC;
- if (now < deadline) {
+ remaining = hrtimer_expires_remaining(&target->obd_recovery_timer);
+ timeout = ktime_divns(remaining, NSEC_PER_SEC);
+ if (timeout > 0) {
+ LCONSOLE_WARN("%s: Client %s (at %s) reconnected, waiting for %d clients in recovery for %lld:%.02lld\n",
+ target->obd_name,
+ obd_uuid2str(&exp->exp_client_uuid),
+ obd_export_nid2str(exp),
+ target->obd_max_recoverable_clients,
+ timeout / 60, timeout % 60);
+ } else {
struct target_distribute_txn_data *tdtd;
int size = 0;
int count = 0;
char *buf = NULL;
- timeout = deadline - now;
tdtd = class_exp2tgt(exp)->lut_tdtd;
if (tdtd && tdtd->tdtd_show_update_logs_retrievers)
buf = tdtd->tdtd_show_update_logs_retrievers(
&size, &count);
if (count > 0)
- LCONSOLE_WARN("%s: Recovery already passed deadline "
- "%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",
- target->obd_name, timeout / 60,
- timeout % 60, count,
- buf ? buf : "unknown (not enough RAM)");
+ LCONSOLE_WARN("%s: Client %s (at %s) reconnecting, waiting for %d MDTs (%s) in recovery for %lld:%.02lld. Please wait until all MDTs recovered or you may force MDT evicition via 'lctl --device %s abort_recovery.\n",
+ target->obd_name,
+ obd_uuid2str(&exp->exp_client_uuid),
+ obd_export_nid2str(exp), count,
+ buf ? buf : "unknown (not enough RAM)",
+ (abs(timeout) + target->obd_recovery_timeout) / 60,
+ (abs(timeout) + target->obd_recovery_timeout) % 60,
+ target->obd_name);
else
- LCONSOLE_WARN("%s: Recovery already passed deadline "
- "%lld:%.02lld. If you do not want to wait "
- "more, please abort the recovery by "
- "force.\n", target->obd_name,
- timeout / 60, timeout % 60);
+ LCONSOLE_WARN("%s: Recovery already passed deadline %lld:%.02lld. If you do not want to wait more, you may force taget eviction via 'lctl --device %s abort_recovery.\n",
+ target->obd_name, abs(timeout) / 60,
+ abs(timeout) % 60, target->obd_name);
if (buf != NULL)
OBD_FREE(buf, size);
- } else {
- timeout = now - deadline;
- LCONSOLE_WARN("%s: Recovery already passed deadline"
- " %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);
}
out_already:
* cause namespace inconsistency */
spin_lock(&export->exp_lock);
export->exp_connecting = 1;
+ export->exp_conn_cnt = 0;
spin_unlock(&export->exp_lock);
conn.cookie = export->exp_handle.h_cookie;
rc = EALREADY;
target->obd_name, cluuid.uuid,
libcfs_nid2str(req->rq_peer.nid),
atomic_read(&export->exp_refcount));
- GOTO(out, rc = -EBUSY);
- } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1) {
- if (!strstr(cluuid.uuid, "mdt"))
- LCONSOLE_WARN("%s: Rejecting reconnect from the "
- "known client %s (at %s) because it "
- "is indicating it is a new client",
- target->obd_name, cluuid.uuid,
- libcfs_nid2str(req->rq_peer.nid));
- GOTO(out, rc = -EALREADY);
- } else {
- OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_RECONNECT, 2 * obd_timeout);
- }
+ GOTO(out, rc = -EBUSY);
+ } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1 &&
+ rc != EALREADY) {
+ if (!strstr(cluuid.uuid, "mdt"))
+ LCONSOLE_WARN("%s: Rejecting reconnect from the "
+ "known client %s (at %s) because it "
+ "is indicating it is a new client",
+ target->obd_name, cluuid.uuid,
+ libcfs_nid2str(req->rq_peer.nid));
+ GOTO(out, rc = -EALREADY);
+ } else {
+ OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_RECONNECT, 2 * obd_timeout);
+ }
if (rc < 0) {
GOTO(out, rc);
/* 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) {
- 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 = 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 %lld:%.02lld\n",
+ struct hrtimer *timer = &target->obd_recovery_timer;
+ ktime_t remaining;
+ s64 timeout, left;
+ int in_progress;
+ int connected;
+ int known;
+ int stale;
+ char *msg;
+
+ connected = atomic_read(&target->obd_connected_clients);
+ in_progress = atomic_read(&target->obd_lock_replay_clients);
+ known = target->obd_max_recoverable_clients;
+ stale = target->obd_stale_clients;
+ remaining = hrtimer_expires_remaining(timer);
+ left = ktime_divns(remaining, NSEC_PER_SEC);
+ if (ktime_to_ns(remaining) > 0) {
+ msg = "to recover in";
+ timeout = left;
+ } else {
+ msg = "already passed deadline";
+ timeout = -left;
+ }
+
+ LCONSOLE_WARN("%s: Denying connection for new client %s (at %s), waiting for %d known clients (%d recovered, %d in progress, and %d evicted) %s %lld:%.02lld\n",
target->obd_name, cluuid.uuid,
- libcfs_nid2str(req->rq_peer.nid), k,
- c - i, i, s, t / 60, t % 60);
+ libcfs_nid2str(req->rq_peer.nid), known,
+ connected - in_progress, in_progress,
+ stale, msg, timeout / 60, timeout % 60);
rc = -EBUSY;
} else {
dont_check_exports:
/* obd_recovery_task_lock should be held */
void target_cancel_recovery_timer(struct obd_device *obd)
{
- CDEBUG(D_HA, "%s: cancel recovery timer\n", obd->obd_name);
- del_timer(&obd->obd_recovery_timer);
+ CDEBUG(D_HA, "%s: cancel recovery timer\n", obd->obd_name);
+ hrtimer_cancel(&obd->obd_recovery_timer);
}
static void target_start_recovery_timer(struct obd_device *obd)
{
+ ktime_t delay;
+
if (obd->obd_recovery_start != 0)
return;
return;
}
- mod_timer(&obd->obd_recovery_timer,
- jiffies + cfs_time_seconds(obd->obd_recovery_timeout));
+ delay = ktime_set(obd->obd_recovery_timeout, 0);
+ hrtimer_start(&obd->obd_recovery_timer, delay, HRTIMER_MODE_REL);
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, time64_t drt,
+static void extend_recovery_timer(struct obd_device *obd, time_t drt,
bool extend)
{
- time64_t now;
- time64_t end;
- time64_t left;
- time64_t to;
+ ktime_t left_ns;
+ time_t left;
+ time_t to;
spin_lock(&obd->obd_dev_lock);
if (!obd->obd_recovering || obd->obd_abort_recovery) {
}
LASSERT(obd->obd_recovery_start != 0);
- now = ktime_get_real_seconds();
to = obd->obd_recovery_timeout;
- end = obd->obd_recovery_start + to;
- left = end - now;
-
- if (extend && (drt > left)) {
- to += drt - left;
- } else if (!extend && (drt > to)) {
- to = drt;
- }
+ left_ns = hrtimer_expires_remaining(&obd->obd_recovery_timer);
+ left = ktime_divns(left_ns, NSEC_PER_SEC);
+ if (extend && (drt > left))
+ to += drt - left;
+ else if (!extend && (drt > to))
+ to = drt;
if (to > obd->obd_recovery_time_hard) {
to = obd->obd_recovery_time_hard;
- CWARN("%s: extended recovery timer reaching hard limit: %lld, extend: %d\n",
+ CWARN("%s: extended recovery timer reaching hard limit: %ld, extend: %d\n",
obd->obd_name, to, extend);
}
if (obd->obd_recovery_timeout < to) {
- obd->obd_recovery_timeout = to;
- end = obd->obd_recovery_start + to;
- mod_timer(&obd->obd_recovery_timer,
- jiffies + cfs_time_seconds(end - now));
- }
+ ktime_t now = ktime_get_real();
+ ktime_t end;
+
+ obd->obd_recovery_timeout = to;
+ end = ktime_set(obd->obd_recovery_start + to, 0);
+ left_ns = ktime_sub(end, now);
+ hrtimer_forward_now(&obd->obd_recovery_timer, left_ns);
+ left = ktime_divns(left_ns, NSEC_PER_MSEC);
+ }
spin_unlock(&obd->obd_dev_lock);
- CDEBUG(D_HA, "%s: recovery timer will expire in %lld seconds\n",
- obd->obd_name, (s64)(end - now));
+ CDEBUG(D_HA, "%s: recovery timer will expire in %ld seconds\n",
+ obd->obd_name, left);
}
/* Reset the timer with each new client connection */
* be extended to make sure the client could be reconnected, in the
* process, the timeout from the new client should be ignored.
*/
-
static void
check_and_start_recovery_timer(struct obd_device *obd,
- struct ptlrpc_request *req,
- int new_client)
+ struct ptlrpc_request *req,
+ int new_client)
{
- time64_t service_time = lustre_msg_get_service_time(req->rq_reqmsg);
- struct obd_device_target *obt = &obd->u.obt;
+ time_t service_time = lustre_msg_get_service_time(req->rq_reqmsg);
+ struct obd_device_target *obt = &obd->u.obt;
- if (!new_client && service_time)
- /* Teach server about old server's estimates, as first guess
- * at how long new requests will take. */
+ if (!new_client && service_time)
+ /* Teach server about old server's estimates, as first guess
+ * at how long new requests will take.
+ */
at_measured(&req->rq_rqbd->rqbd_svcpt->scp_at_estimate,
- service_time);
+ service_time);
- target_start_recovery_timer(obd);
+ target_start_recovery_timer(obd);
/* Convert the service time to RPC timeout,
* and reuse service_time to limit stack usage.
(void)handler(req);
lu_context_exit(&thread->t_env->le_ctx);
- /* don't reset timer for final stage */
- if (!exp_finished(req->rq_export)) {
- time64_t to = obd_timeout;
+ /* don't reset timer for final stage */
+ if (!exp_finished(req->rq_export)) {
+ time_t to = obd_timeout;
- /**
- * Add request timeout to the recovery time so next request from
- * this client may come in recovery time
- */
- if (!AT_OFF) {
+ /**
+ * Add request timeout to the recovery time so next request from
+ * this client may come in recovery time
+ */
+ if (!AT_OFF) {
struct ptlrpc_service_part *svcpt;
svcpt = req->rq_rqbd->rqbd_svcpt;
* the client will recalculate the timeout according to
* current server estimate service time, so we will
* use the maxium timeout here for waiting the client
- * sending the next req */
- to = max((int)at_est2timeout(
- at_get(&svcpt->scp_at_estimate)),
- (int)lustre_msg_get_timeout(req->rq_reqmsg));
+ * sending the next req
+ */
+ to = max_t(time_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)),
+ lustre_msg_get_timeout(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
RETURN(rc);
}
- thread->t_env = env;
- thread->t_id = -1; /* force filter_iobuf_get/put to use local buffers */
- env->le_ctx.lc_thread = thread;
+ thread->t_env = env;
+ thread->t_id = -1; /* force filter_iobuf_get/put to use local buffers */
+ env->le_ctx.lc_thread = thread;
tgt_io_thread_init(thread); /* init thread_big_cache for IO requests */
- thread->t_watchdog = NULL;
CDEBUG(D_HA, "%s: started recovery thread pid %d\n", obd->obd_name,
current_pid());
}
EXPORT_SYMBOL(target_recovery_fini);
-static void target_recovery_expired(unsigned long castmeharder)
+static enum hrtimer_restart target_recovery_expired(struct hrtimer *timer)
{
- struct obd_device *obd = (struct obd_device *)castmeharder;
- CDEBUG(D_HA, "%s: recovery timed out; %d clients are still in recovery"
- " after %llus (%d clients connected)\n",
+ struct obd_device *obd = container_of(timer, struct obd_device,
+ obd_recovery_timer);
+
+ CDEBUG(D_HA,
+ "%s: recovery timed out; %d clients are still in recovery after %llu seconds (%d clients connected)\n",
obd->obd_name, atomic_read(&obd->obd_lock_replay_clients),
- (s64)(ktime_get_real_seconds() - obd->obd_recovery_start),
+ ktime_get_real_seconds() - obd->obd_recovery_start,
atomic_read(&obd->obd_connected_clients));
obd->obd_recovery_expired = 1;
wake_up(&obd->obd_next_transno_waitq);
+ return HRTIMER_NORESTART;
}
void target_recovery_init(struct lu_target *lut, svc_handler_t handler)
CDEBUG(D_HA, "RECOVERY: service %s, %d recoverable clients, "
"last_transno %llu\n", obd->obd_name,
obd->obd_max_recoverable_clients, obd->obd_last_committed);
- LASSERT(obd->obd_stopping == 0);
- obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
- obd->obd_recovery_start = 0;
- obd->obd_recovery_end = 0;
-
- setup_timer(&obd->obd_recovery_timer, target_recovery_expired,
- (unsigned long)obd);
+ LASSERT(obd->obd_stopping == 0);
+ obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
+ obd->obd_recovery_start = 0;
+ obd->obd_recovery_end = 0;
+
+ hrtimer_init(&obd->obd_recovery_timer, CLOCK_REALTIME,
+ HRTIMER_MODE_REL);
+ obd->obd_recovery_timer.function = &target_recovery_expired;
target_start_recovery_thread(lut, handler);
}
EXPORT_SYMBOL(target_recovery_init);
RETURN(0);
}
-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);
-}
-
void target_committed_to_req(struct ptlrpc_request *req)
{
struct obd_export *exp = req->rq_export;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff