X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lib.c;h=2b62e25f887e0d1bb0c65ee12e01bf83e3423cd5;hp=feacd50188a6d0cbb7e06f8839cb93d48f01c473;hb=37b29a8f709aa4bd2e22cbb3549257b28cd85610;hpb=0098396983e1075668414aa5298a4990e61ffbda diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c index feacd50..2b62e25f 100644 --- a/lustre/ldlm/ldlm_lib.c +++ b/lustre/ldlm/ldlm_lib.c @@ -271,13 +271,13 @@ static int osc_on_mdt(char *obdname) * 3 - inactive-on-startup * 4 - restrictive net */ -int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) +int client_obd_setup(struct obd_device *obd, struct lustre_cfg *lcfg) { - struct client_obd *cli = &obddev->u.cli; + struct client_obd *cli = &obd->u.cli; struct obd_import *imp; struct obd_uuid server_uuid; int rq_portal, rp_portal, connect_op; - const char *name = obddev->obd_type->typ_name; + const char *name = obd->obd_type->typ_name; enum ldlm_ns_type ns_type = LDLM_NS_TYPE_UNKNOWN; char *cli_name = lustre_cfg_buf(lcfg, 0); struct ptlrpc_connection fake_conn = { .c_self = 0, @@ -378,7 +378,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) * ptlrpc_connect_interpret(). */ client_adjust_max_dirty(cli); - INIT_LIST_HEAD(&cli->cl_cache_waiters); + init_waitqueue_head(&cli->cl_cache_waiters); INIT_LIST_HEAD(&cli->cl_loi_ready_list); INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list); INIT_LIST_HEAD(&cli->cl_loi_write_list); @@ -451,7 +451,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) } else if (cfs_totalram_pages() >> (20 - PAGE_SHIFT) <= 512 /* MB */) { cli->cl_max_rpcs_in_flight = 4; } else { - if (osc_on_mdt(obddev->obd_name)) + if (osc_on_mdt(obd->obd_name)) cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_MAX; else cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_DEFAULT; @@ -482,12 +482,12 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) } ptlrpc_init_client(rq_portal, rp_portal, name, - &obddev->obd_ldlm_client); + &obd->obd_ldlm_client); - imp = class_new_import(obddev); + imp = class_new_import(obd); if (imp == NULL) GOTO(err_ldlm, rc = -ENOENT); - imp->imp_client = &obddev->obd_ldlm_client; + imp->imp_client = &obd->obd_ldlm_client; imp->imp_connect_op = connect_op; memcpy(cli->cl_target_uuid.uuid, lustre_cfg_buf(lcfg, 1), LUSTRE_CFG_BUFLEN(lcfg, 1)); @@ -499,7 +499,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) if (refnet == LNET_NIDNET(LNET_NID_ANY)) { rc = -EINVAL; CERROR("%s: bad mount option 'network=%s': rc = %d\n", - obddev->obd_name, lustre_cfg_string(lcfg, 4), + obd->obd_name, lustre_cfg_string(lcfg, 4), rc); GOTO(err_import, rc); } @@ -521,7 +521,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) { if (!strcmp(lustre_cfg_string(lcfg, 3), "inactive")) { CDEBUG(D_HA, "marking %s %s->%s as inactive\n", - name, obddev->obd_name, + name, obd->obd_name, cli->cl_target_uuid.uuid); spin_lock(&imp->imp_lock); imp->imp_deactive = 1; @@ -529,13 +529,13 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) } } - obddev->obd_namespace = ldlm_namespace_new(obddev, obddev->obd_name, - LDLM_NAMESPACE_CLIENT, - LDLM_NAMESPACE_GREEDY, - ns_type); - if (obddev->obd_namespace == NULL) { + obd->obd_namespace = ldlm_namespace_new(obd, obd->obd_name, + LDLM_NAMESPACE_CLIENT, + LDLM_NAMESPACE_GREEDY, + ns_type); + if (obd->obd_namespace == NULL) { CERROR("Unable to create client namespace - %s\n", - obddev->obd_name); + obd->obd_name); GOTO(err_import, rc = -ENOMEM); } @@ -555,17 +555,17 @@ err: } EXPORT_SYMBOL(client_obd_setup); -int client_obd_cleanup(struct obd_device *obddev) +int client_obd_cleanup(struct obd_device *obd) { - struct client_obd *cli = &obddev->u.cli; + struct client_obd *cli = &obd->u.cli; ENTRY; - ldlm_namespace_free_post(obddev->obd_namespace); - obddev->obd_namespace = NULL; + ldlm_namespace_free_post(obd->obd_namespace); + obd->obd_namespace = NULL; - obd_cleanup_client_import(obddev); - LASSERT(obddev->u.cli.cl_import == NULL); + obd_cleanup_client_import(obd); + LASSERT(obd->u.cli.cl_import == NULL); ldlm_put_ref(); @@ -787,12 +787,37 @@ int server_disconnect_export(struct obd_export *exp) } EXPORT_SYMBOL(server_disconnect_export); +static inline int target_check_recovery_timer(struct obd_device *target) +{ + ktime_t remaining; + s64 timeout; + + if (!target->obd_recovering || target->obd_recovery_start == 0) + return 0; + + remaining = hrtimer_expires_remaining(&target->obd_recovery_timer); + timeout = ktime_divns(remaining, NSEC_PER_SEC); + if (timeout > -30) + return 0; + + /* the recovery timer should expire, but it isn't triggered, + * it's better to abort the recovery of this target to speed up + * the recovery of the whole cluster. */ + spin_lock(&target->obd_dev_lock); + if (target->obd_recovering) { + CERROR("%s: Aborting recovery\n", target->obd_name); + target->obd_abort_recovery = 1; + wake_up(&target->obd_next_transno_waitq); + } + spin_unlock(&target->obd_dev_lock); + return 0; +} + /* * -------------------------------------------------------------------------- * from old lib/target.c * -------------------------------------------------------------------------- */ - static int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp, struct obd_uuid *cluuid) @@ -853,6 +878,8 @@ static int target_handle_reconnect(struct lustre_handle *conn, int count = 0; char *buf = NULL; + target_check_recovery_timer(target); + tdtd = class_exp2tgt(exp)->lut_tdtd; if (tdtd && tdtd->tdtd_show_update_logs_retrievers) buf = tdtd->tdtd_show_update_logs_retrievers( @@ -1138,8 +1165,8 @@ int target_handle_connect(struct ptlrpc_request *req) OBD_CONNECT_MDS_MDS) != 0; /* - * OBD_CONNECT_MNE_SWAB is defined as OBD_CONNECT_MDS_MDS - * for Imperative Recovery connection from MGC to MGS. + * OBD_CONNECT_MNE_SWAB is removed at 2.14 + * Checking OBD_CONNECT_FID can be removed in the future. * * Via check OBD_CONNECT_FID, we can distinguish whether * the OBD_CONNECT_MDS_MDS/OBD_CONNECT_MNE_SWAB is from @@ -1340,6 +1367,8 @@ no_export: } else { msg = "already passed deadline"; timeout = -left; + + target_check_recovery_timer(target); } 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", @@ -1418,11 +1447,8 @@ dont_check_exports: 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) && - !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); + if (export->exp_connection->c_peer.nid != req->rq_peer.nid) + obd_nid_del(export->exp_obd, export); ptlrpc_connection_put(export->exp_connection); } @@ -1430,10 +1456,10 @@ dont_check_exports: export->exp_connection = ptlrpc_connection_get(req->rq_peer, req->rq_self, &cluuid); - if (hlist_unhashed(&export->exp_nid_hash)) - cfs_hash_add(export->exp_obd->obd_nid_hash, - &export->exp_connection->c_peer.nid, - &export->exp_nid_hash); + if (!export->exp_connection) + GOTO(out, rc = -ENOTCONN); + + obd_nid_add(export->exp_obd, export); lustre_msg_set_handle(req->rq_repmsg, &conn); @@ -1647,8 +1673,8 @@ static void target_finish_recovery(struct lu_target *lut) elapsed_time = max_t(time64_t, now - obd->obd_recovery_start, 1); LCONSOLE_INFO("%s: Recovery over after %lld:%.02lld, of %d clients %d recovered and %d %s evicted.\n", - obd->obd_name, (s64)elapsed_time / 60, - (s64)elapsed_time % 60, + obd->obd_name, elapsed_time / 60, + elapsed_time % 60, atomic_read(&obd->obd_max_recoverable_clients), atomic_read(&obd->obd_connected_clients), obd->obd_stale_clients, @@ -1805,7 +1831,7 @@ static void target_start_recovery_timer(struct obd_device *obd) hrtimer_start(&obd->obd_recovery_timer, delay, HRTIMER_MODE_ABS); spin_unlock(&obd->obd_dev_lock); - LCONSOLE_WARN("%s: Will be in recovery for at least %lu:%02lu, or until %d client%s reconnect%s\n", + LCONSOLE_WARN("%s: Will be in recovery for at least %u:%02u, or until %d client%s reconnect%s\n", obd->obd_name, obd->obd_recovery_timeout / 60, obd->obd_recovery_timeout % 60, @@ -1823,12 +1849,12 @@ static void target_start_recovery_timer(struct obd_device *obd) * at least; otherwise, make sure the recovery timeout value is not less * than @dr_timeout. */ -static void extend_recovery_timer(struct obd_device *obd, time_t dr_timeout, +static void extend_recovery_timer(struct obd_device *obd, timeout_t dr_timeout, bool extend) { ktime_t left_ns; - time_t timeout; - time_t left; + timeout_t timeout; + timeout_t left; spin_lock(&obd->obd_dev_lock); if (!obd->obd_recovering || obd->obd_abort_recovery || @@ -1849,16 +1875,17 @@ static void extend_recovery_timer(struct obd_device *obd, time_t dr_timeout, */ if (dr_timeout > left) { timeout += dr_timeout - left; - timeout = min_t(time_t, obd->obd_recovery_time_hard, + timeout = min_t(timeout_t, obd->obd_recovery_time_hard, timeout); } } else { - timeout = clamp_t(time_t, dr_timeout, obd->obd_recovery_timeout, + timeout = clamp_t(timeout_t, dr_timeout, + obd->obd_recovery_timeout, obd->obd_recovery_time_hard); } if (timeout == obd->obd_recovery_time_hard) - CWARN("%s: extended recovery timer reached hard limit: %ld, extend: %d\n", + CWARN("%s: extended recovery timer reached hard limit: %d, extend: %d\n", obd->obd_name, timeout, extend); if (obd->obd_recovery_timeout < timeout) { @@ -1873,7 +1900,7 @@ static void extend_recovery_timer(struct obd_device *obd, time_t dr_timeout, } spin_unlock(&obd->obd_dev_lock); - CDEBUG(D_HA, "%s: recovery timer will expire in %ld seconds\n", + CDEBUG(D_HA, "%s: recovery timer will expire in %d seconds\n", obd->obd_name, left); } @@ -1892,40 +1919,40 @@ check_and_start_recovery_timer(struct obd_device *obd, struct ptlrpc_request *req, int new_client) { - time_t service_time = lustre_msg_get_service_time(req->rq_reqmsg); + timeout_t service_timeout = lustre_msg_get_service_timeout(req->rq_reqmsg); struct obd_device_target *obt = &obd->u.obt; - if (!new_client && service_time) + if (!new_client && service_timeout) /* * 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_timeout); target_start_recovery_timer(obd); /* * Convert the service time to RPC timeout, - * and reuse service_time to limit stack usage. + * and reuse service_timeout to limit stack usage. */ - service_time = at_est2timeout(service_time); + service_timeout = at_est2timeout(service_timeout); if (OBD_FAIL_CHECK(OBD_FAIL_TGT_SLUGGISH_NET) && - service_time < at_extra) - service_time = at_extra; + service_timeout < at_extra) + service_timeout = at_extra; /* - * We expect other clients to timeout within service_time, then try + * We expect other clients to timeout within service_timeout, then try * to reconnect, then try the failover server. The max delay between * connect attempts is SWITCH_MAX + SWITCH_INC + INITIAL. */ - service_time += 2 * INITIAL_CONNECT_TIMEOUT; + service_timeout += 2 * INITIAL_CONNECT_TIMEOUT; LASSERT(obt->obt_magic == OBT_MAGIC); - service_time += 2 * (CONNECTION_SWITCH_MAX + CONNECTION_SWITCH_INC); - if (service_time > obd->obd_recovery_timeout && !new_client) - extend_recovery_timer(obd, service_time, false); + service_timeout += 2 * (CONNECTION_SWITCH_MAX + CONNECTION_SWITCH_INC); + if (service_timeout > obd->obd_recovery_timeout && !new_client) + extend_recovery_timer(obd, service_timeout, false); } /** Health checking routines */ @@ -2277,14 +2304,15 @@ static void handle_recovery_req(struct ptlrpc_thread *thread, /* don't reset timer for final stage */ if (!exp_finished(req->rq_export)) { - time_t to = obd_timeout; + timeout_t timeout = obd_timeout; /** - * Add request timeout to the recovery time so next request from + * 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; + timeout_t est_timeout; svcpt = req->rq_rqbd->rqbd_svcpt; /* @@ -2294,18 +2322,19 @@ static void handle_recovery_req(struct ptlrpc_thread *thread, * use the maxium timeout here for waiting the client * sending the next req */ - to = max_t(time_t, - at_est2timeout(at_get(&svcpt->scp_at_estimate)), - lustre_msg_get_timeout(req->rq_reqmsg)); + est_timeout = at_get(&svcpt->scp_at_estimate); + timeout = max_t(timeout_t, at_est2timeout(est_timeout), + 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 * (see ptlrpc_replay_req) */ - to += 2 * lustre_msg_get_service_time(req->rq_reqmsg); + timeout += 2 * lustre_msg_get_service_timeout(req->rq_reqmsg); } - extend_recovery_timer(class_exp2obd(req->rq_export), to, true); + extend_recovery_timer(class_exp2obd(req->rq_export), timeout, + true); } EXIT; } @@ -2340,7 +2369,7 @@ static int check_for_recovery_ready(struct lu_target *lut) extend_recovery_timer(obd, obd->obd_recovery_timeout, true); CDEBUG(D_HA, - "%s update recovery is not ready, extend recovery %lu\n", + "%s update recovery is not ready, extend recovery %d\n", obd->obd_name, obd->obd_recovery_timeout); return 0; } @@ -2507,7 +2536,7 @@ static void replay_request_or_update(struct lu_env *env, continue; } - LASSERT(trd->trd_processing_task == current_pid()); + LASSERT(trd->trd_processing_task == current->pid); DEBUG_REQ(D_HA, req, "processing x%llu t%lld from %s", req->rq_xid, lustre_msg_get_transno(req->rq_reqmsg), @@ -2608,8 +2637,8 @@ static int target_recovery_thread(void *arg) tgt_io_thread_init(thread); /* init thread_big_cache for IO requests */ CDEBUG(D_HA, "%s: started recovery thread pid %d\n", obd->obd_name, - current_pid()); - trd->trd_processing_task = current_pid(); + current->pid); + trd->trd_processing_task = current->pid; spin_lock(&obd->obd_dev_lock); obd->obd_recovering = 1; @@ -2638,7 +2667,7 @@ static int target_recovery_thread(void *arg) CDEBUG(D_INFO, "2: lock replay stage - %d clients\n", atomic_read(&obd->obd_lock_replay_clients)); while ((req = target_next_replay_lock(lut))) { - LASSERT(trd->trd_processing_task == current_pid()); + LASSERT(trd->trd_processing_task == current->pid); DEBUG_REQ(D_HA, req, "processing lock from %s:", libcfs_nid2str(req->rq_peer.nid)); handle_recovery_req(thread, req, @@ -2666,7 +2695,7 @@ static int target_recovery_thread(void *arg) target_cancel_recovery_timer(obd); spin_unlock(&obd->obd_recovery_task_lock); while ((req = target_next_final_ping(obd))) { - LASSERT(trd->trd_processing_task == current_pid()); + LASSERT(trd->trd_processing_task == current->pid); DEBUG_REQ(D_HA, req, "processing final ping from %s:", libcfs_nid2str(req->rq_peer.nid)); handle_recovery_req(thread, req, @@ -2763,7 +2792,7 @@ static enum hrtimer_restart target_recovery_expired(struct hrtimer *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), - ktime_get_real_seconds() - obd->obd_recovery_start, + ktime_get_seconds() - obd->obd_recovery_start, atomic_read(&obd->obd_connected_clients)); obd->obd_recovery_expired = 1; @@ -2847,7 +2876,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req, ENTRY; - if (obd->obd_recovery_data.trd_processing_task == current_pid()) { + if (obd->obd_recovery_data.trd_processing_task == current->pid) { /* Processing the queue right now, don't re-add. */ RETURN(1); } @@ -2861,8 +2890,8 @@ int target_queue_recovery_request(struct ptlrpc_request *req, cfs_fail_val = 0; wake_up(&cfs_race_waitq); - set_current_state(TASK_INTERRUPTIBLE); - schedule_timeout(cfs_time_seconds(1)); + schedule_timeout_interruptible( + cfs_time_seconds(1)); } }