X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lib.c;h=b0649fd0cf745e18e1f2ef38ba210ceec0e21bfe;hp=268f77abbaf679c2a0c7118a457ebd7c5f67a013;hb=580ef453d1d6f901c3227a037137a7db9e600009;hpb=d9d48aab9e3d4e00037598be5427dd196cccaad7 diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c index 268f77a..b0649fd 100644 --- a/lustre/ldlm/ldlm_lib.c +++ b/lustre/ldlm/ldlm_lib.c @@ -90,9 +90,8 @@ static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid, list_for_each_entry(item, &imp->imp_conn_list, oic_item) { if (obd_uuid_equals(uuid, &item->oic_uuid)) { if (priority) { - list_del(&item->oic_item); - list_add(&item->oic_item, - &imp->imp_conn_list); + list_move(&item->oic_item, + &imp->imp_conn_list); item->oic_last_attempt = 0; } CDEBUG(D_HA, "imp %p@%s: found existing conn %s%s\n", @@ -272,13 +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, @@ -379,7 +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); @@ -414,7 +413,9 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) init_waitqueue_head(&cli->cl_destroy_waitq); atomic_set(&cli->cl_destroy_in_flight, 0); + cli->cl_supp_cksum_types = OBD_CKSUM_CRC32; + cli->cl_preferred_cksum_type = 0; #ifdef ENABLE_CHECKSUM /* Turn on checksumming by default. */ cli->cl_checksum = 1; @@ -433,7 +434,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) */ cli->cl_max_pages_per_rpc = PTLRPC_MAX_BRW_PAGES; - cli->cl_max_short_io_bytes = OBD_MAX_SHORT_IO_BYTES; + cli->cl_max_short_io_bytes = OBD_DEF_SHORT_IO_BYTES; /* * set cl_chunkbits default value to PAGE_SHIFT, @@ -443,14 +444,14 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) if (!strcmp(name, LUSTRE_MDC_NAME)) { cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_DEFAULT; - } else if (totalram_pages >> (20 - PAGE_SHIFT) <= 128 /* MB */) { + } else if (cfs_totalram_pages() >> (20 - PAGE_SHIFT) <= 128 /* MB */) { cli->cl_max_rpcs_in_flight = 2; - } else if (totalram_pages >> (20 - PAGE_SHIFT) <= 256 /* MB */) { + } else if (cfs_totalram_pages() >> (20 - PAGE_SHIFT) <= 256 /* MB */) { cli->cl_max_rpcs_in_flight = 3; - } else if (totalram_pages >> (20 - PAGE_SHIFT) <= 512 /* MB */) { + } 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; @@ -481,12 +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)); @@ -498,7 +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); } @@ -520,7 +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; @@ -528,13 +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); } @@ -554,17 +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(); @@ -786,12 +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) @@ -844,7 +870,7 @@ static int target_handle_reconnect(struct lustre_handle *conn, target->obd_name, obd_uuid2str(&exp->exp_client_uuid), obd_export_nid2str(exp), - target->obd_max_recoverable_clients, + atomic_read(&target->obd_max_recoverable_clients), timeout / 60, timeout % 60); } else { struct target_distribute_txn_data *tdtd; @@ -852,6 +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( @@ -1086,6 +1114,8 @@ int target_handle_connect(struct ptlrpc_request *req) size = req_capsule_get_size(&req->rq_pill, &RMF_CONNECT_DATA, RCL_CLIENT); + if (size < 0 || size > 8 * sizeof(struct obd_connect_data)) + GOTO(out, rc = -EPROTO); data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA); if (!data) GOTO(out, rc = -EPROTO); @@ -1135,8 +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 @@ -1173,7 +1203,7 @@ int target_handle_connect(struct ptlrpc_request *req) if (obd_uuid_equals(&cluuid, &target->obd_uuid)) goto dont_check_exports; - export = cfs_hash_lookup(target->obd_uuid_hash, &cluuid); + export = obd_uuid_lookup(target, &cluuid); if (!export) goto no_export; @@ -1241,6 +1271,10 @@ int target_handle_connect(struct ptlrpc_request *req) rc = -EALREADY; class_export_put(export); export = NULL; + } else if (OBD_FAIL_PRECHECK(OBD_FAIL_TGT_RECOVERY_CONNECT) && + !lw_client) { + spin_unlock(&export->exp_lock); + rc = -EAGAIN; } else { export->exp_connecting = 1; spin_unlock(&export->exp_lock); @@ -1258,7 +1292,7 @@ no_export: LCONSOLE_WARN("%s: Client %s (at %s) refused connection, still busy with %d references\n", target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), - atomic_read(&export->exp_refcount)); + refcount_read(&export->exp_handle.h_ref)); GOTO(out, rc = -EBUSY); } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1 && rc != EALREADY) { @@ -1277,7 +1311,7 @@ no_export: 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, ktime_get_seconds(), export ? export->exp_last_request_time : 0); /* @@ -1321,7 +1355,8 @@ no_export: connected = atomic_read(&target->obd_connected_clients); in_progress = atomic_read(&target->obd_lock_replay_clients); - known = target->obd_max_recoverable_clients; + known = + atomic_read(&target->obd_max_recoverable_clients); stale = target->obd_stale_clients; remaining = hrtimer_expires_remaining(timer); left = ktime_divns(remaining, NSEC_PER_SEC); @@ -1332,6 +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", @@ -1410,11 +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); } @@ -1422,10 +1456,7 @@ 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); + obd_nid_add(export->exp_obd, export); lustre_msg_set_handle(req->rq_repmsg, &conn); @@ -1477,9 +1508,10 @@ dont_check_exports: * condition. */ if (new_mds_mds_conn) - target->obd_max_recoverable_clients++; + atomic_inc(&target->obd_max_recoverable_clients); + if (atomic_inc_return(&target->obd_connected_clients) == - target->obd_max_recoverable_clients) + atomic_read(&target->obd_max_recoverable_clients)) wake_up(&target->obd_next_transno_waitq); } @@ -1632,15 +1664,15 @@ static void target_finish_recovery(struct lu_target *lut) /* Only log a recovery message when recovery has occurred. */ if (obd->obd_recovery_start) { - time64_t now = ktime_get_real_seconds(); + time64_t now = ktime_get_seconds(); time64_t elapsed_time; 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_max_recoverable_clients, + 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, obd->obd_stale_clients == 1 ? "was" : "were"); @@ -1663,7 +1695,7 @@ static void target_finish_recovery(struct lu_target *lut) } spin_unlock(&obd->obd_recovery_task_lock); - obd->obd_recovery_end = ktime_get_real_seconds(); + obd->obd_recovery_end = ktime_get_seconds(); /* When recovery finished, cleanup orphans on MDS and OST. */ if (obd->obd_type && OBP(obd, postrecov)) { @@ -1679,9 +1711,8 @@ static void target_finish_recovery(struct lu_target *lut) static void abort_req_replay_queue(struct obd_device *obd) { struct ptlrpc_request *req, *n; - struct list_head abort_list; + LIST_HEAD(abort_list); - INIT_LIST_HEAD(&abort_list); spin_lock(&obd->obd_recovery_task_lock); list_splice_init(&obd->obd_req_replay_queue, &abort_list); spin_unlock(&obd->obd_recovery_task_lock); @@ -1700,9 +1731,8 @@ static void abort_req_replay_queue(struct obd_device *obd) static void abort_lock_replay_queue(struct obd_device *obd) { struct ptlrpc_request *req, *n; - struct list_head abort_list; + LIST_HEAD(abort_list); - INIT_LIST_HEAD(&abort_list); spin_lock(&obd->obd_recovery_task_lock); list_splice_init(&obd->obd_lock_replay_queue, &abort_list); spin_unlock(&obd->obd_recovery_task_lock); @@ -1729,9 +1759,8 @@ static void abort_lock_replay_queue(struct obd_device *obd) void target_cleanup_recovery(struct obd_device *obd) { struct ptlrpc_request *req, *n; - struct list_head clean_list; + LIST_HEAD(clean_list); - INIT_LIST_HEAD(&clean_list); spin_lock(&obd->obd_dev_lock); if (!obd->obd_recovering) { spin_unlock(&obd->obd_dev_lock); @@ -1793,18 +1822,21 @@ static void target_start_recovery_timer(struct obd_device *obd) return; } - 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(); + obd->obd_recovery_start = ktime_get_seconds(); + delay = ktime_set(obd->obd_recovery_start + + obd->obd_recovery_timeout, 0); + 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 %llu:%02llu, 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, - obd->obd_max_recoverable_clients, - (obd->obd_max_recoverable_clients == 1) ? "" : "s", - (obd->obd_max_recoverable_clients == 1) ? "s" : ""); + atomic_read(&obd->obd_max_recoverable_clients), + (atomic_read(&obd->obd_max_recoverable_clients) == 1) ? + "" : "s", + (atomic_read(&obd->obd_max_recoverable_clients) == 1) ? + "s" : ""); } /** @@ -1814,15 +1846,16 @@ 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) { + if (!obd->obd_recovering || obd->obd_abort_recovery || + obd->obd_stopping) { spin_unlock(&obd->obd_dev_lock); return; } @@ -1839,31 +1872,32 @@ 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) { - ktime_t now = ktime_get_real(); - ktime_t end; + ktime_t end, now; obd->obd_recovery_timeout = timeout; end = ktime_set(obd->obd_recovery_start + timeout, 0); + now = ktime_set(ktime_get_seconds(), 0); left_ns = ktime_sub(end, now); - hrtimer_forward_now(&obd->obd_recovery_timer, left_ns); - left = ktime_divns(left_ns, NSEC_PER_MSEC); + hrtimer_start(&obd->obd_recovery_timer, end, HRTIMER_MODE_ABS); + left = ktime_divns(left_ns, NSEC_PER_SEC); } 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); } @@ -1882,40 +1916,40 @@ 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 */ @@ -1990,7 +2024,8 @@ static int check_for_next_transno(struct lu_target *lut) CDEBUG(D_HA, "max: %d, connected: %d, completed: %d, queue_len: %d, req_transno: %llu, next_transno: %llu\n", - obd->obd_max_recoverable_clients, connected, completed, + atomic_read(&obd->obd_max_recoverable_clients), + connected, completed, queue_len, req_transno, next_transno); if (obd->obd_abort_recovery) { @@ -2084,7 +2119,7 @@ repeat: last = now; } } - if (obd->obd_recovery_start != 0 && ktime_get_real_seconds() >= + if (obd->obd_recovery_start != 0 && ktime_get_seconds() >= (obd->obd_recovery_start + obd->obd_recovery_time_hard)) { __u64 next_update_transno = 0; @@ -2104,13 +2139,13 @@ repeat: * yet, let's wait those threads stopped */ if (next_update_transno == 0) { - struct l_wait_info lwi = { 0 }; - - l_wait_event(tdtd->tdtd_recovery_threads_waitq, - atomic_read( - &tdtd->tdtd_recovery_threads_count) == 0, - &lwi); + spin_unlock(&obd->obd_recovery_task_lock); + wait_event_idle( + tdtd->tdtd_recovery_threads_waitq, + atomic_read(&tdtd->tdtd_recovery_threads_count) + == 0); + spin_lock(&obd->obd_recovery_task_lock); next_update_transno = distribute_txn_get_next_transno( lut->lut_tdtd); @@ -2144,22 +2179,21 @@ repeat: while (wait_event_timeout(obd->obd_next_transno_waitq, check_routine(lut), - msecs_to_jiffies(60 * MSEC_PER_SEC)) == 0) + cfs_time_seconds(60)) == 0) ; /* wait indefinitely for event, but don't trigger watchdog */ if (obd->obd_abort_recovery) { CWARN("recovery is aborted, evict exports in recovery\n"); if (lut->lut_tdtd != NULL) { - struct l_wait_info lwi = { 0 }; - tdtd = lut->lut_tdtd; /* * Let's wait all of the update log recovery thread * finished */ - l_wait_event(tdtd->tdtd_recovery_threads_waitq, - atomic_read(&tdtd->tdtd_recovery_threads_count) == 0, - &lwi); + wait_event_idle( + tdtd->tdtd_recovery_threads_waitq, + atomic_read(&tdtd->tdtd_recovery_threads_count) + == 0); /* Then abort the update recovery list */ dtrq_list_destroy(lut->lut_tdtd); } @@ -2167,8 +2201,10 @@ repeat: /** evict exports which didn't finish recovery yet */ class_disconnect_stale_exports(obd, exp_finished); return 1; - } else if (obd->obd_recovery_expired) { + } else if (obd->obd_recovery_expired && + obd->obd_recovery_timeout < obd->obd_recovery_time_hard) { obd->obd_recovery_expired = 0; + /** If some clients died being recovered, evict them */ LCONSOLE_WARN("%s: recovery is timed out, evict stale exports\n", obd->obd_name); @@ -2261,16 +2297,19 @@ static void handle_recovery_req(struct ptlrpc_thread *thread, (void)handler(req); lu_context_exit(&thread->t_env->le_ctx); + req->rq_svc_thread->t_env->le_ses = NULL; + /* 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; /* @@ -2280,18 +2319,19 @@ 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; } @@ -2304,19 +2344,21 @@ static int check_for_recovery_ready(struct lu_target *lut) CDEBUG(D_HA, "connected %d stale %d max_recoverable_clients %d abort %d expired %d\n", - clnts, obd->obd_stale_clients, obd->obd_max_recoverable_clients, + clnts, obd->obd_stale_clients, + atomic_read(&obd->obd_max_recoverable_clients), obd->obd_abort_recovery, obd->obd_recovery_expired); if (!obd->obd_abort_recovery && !obd->obd_recovery_expired) { - LASSERT(clnts <= obd->obd_max_recoverable_clients); + LASSERT(clnts <= + atomic_read(&obd->obd_max_recoverable_clients)); if (clnts + obd->obd_stale_clients < - obd->obd_max_recoverable_clients) + atomic_read(&obd->obd_max_recoverable_clients)) return 0; } if (lut->lut_tdtd != NULL) { if (!lut->lut_tdtd->tdtd_replay_ready && - !obd->obd_abort_recovery) { + !obd->obd_abort_recovery && !obd->obd_stopping) { /* * Let's extend recovery timer, in case the recovery * timer expired, and some clients got evicted @@ -2324,7 +2366,7 @@ 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 %llu\n", + "%s update recovery is not ready, extend recovery %d\n", obd->obd_name, obd->obd_recovery_timeout); return 0; } @@ -2396,7 +2438,7 @@ static void drop_duplicate_replay_req(struct lu_env *env, struct ptlrpc_request *req) { DEBUG_REQ(D_HA, req, - "remove t%lld from %s because of duplicate update records are found.\n", + "remove t%lld from %s because duplicate update records found", lustre_msg_get_transno(req->rq_reqmsg), libcfs_nid2str(req->rq_peer.nid)); @@ -2412,7 +2454,7 @@ static void drop_duplicate_replay_req(struct lu_env *env, lustre_msg_set_transno(req->rq_repmsg, req->rq_transno); target_send_reply(req, req->rq_status, 0); } else { - DEBUG_REQ(D_ERROR, req, "wrong opc from %s\n", + DEBUG_REQ(D_ERROR, req, "wrong opc from %s", libcfs_nid2str(req->rq_peer.nid)); } target_exp_dequeue_req_replay(req); @@ -2491,7 +2533,7 @@ 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), @@ -2592,8 +2634,8 @@ 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; @@ -2622,8 +2664,8 @@ 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()); - DEBUG_REQ(D_HA, req, "processing lock from %s: ", + 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, trd->trd_recovery_handler); @@ -2650,8 +2692,8 @@ 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()); - DEBUG_REQ(D_HA, req, "processing final ping from %s: ", + 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, trd->trd_recovery_handler); @@ -2747,7 +2789,7 @@ 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; @@ -2762,7 +2804,7 @@ void target_recovery_init(struct lu_target *lut, svc_handler_t handler) if (lut->lut_bottom->dd_rdonly) return; - if (obd->obd_max_recoverable_clients == 0) { + if (atomic_read(&obd->obd_max_recoverable_clients) == 0) { /** Update server last boot epoch */ tgt_boot_epoch_update(lut); return; @@ -2770,14 +2812,15 @@ 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); + atomic_read(&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; - hrtimer_init(&obd->obd_recovery_timer, CLOCK_REALTIME, - HRTIMER_MODE_REL); + hrtimer_init(&obd->obd_recovery_timer, CLOCK_MONOTONIC, + HRTIMER_MODE_ABS); obd->obd_recovery_timer.function = &target_recovery_expired; target_start_recovery_thread(lut, handler); } @@ -2830,7 +2873,7 @@ 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); } @@ -2844,8 +2887,8 @@ 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)); } } @@ -2891,7 +2934,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req, if (duplicate != NULL) { DEBUG_REQ(D_HA, duplicate, - "put prev final req\n"); + "put prev final req"); target_request_copy_put(duplicate); } RETURN(0); @@ -3286,16 +3329,6 @@ void ldlm_dump_export_locks(struct obd_export *exp) #endif #ifdef HAVE_SERVER_SUPPORT -static int target_bulk_timeout(void *data) -{ - ENTRY; - /* - * We don't fail the connection here, because having the export - * killed makes the (vital) call to commitrw very sad. - */ - RETURN(1); -} - static inline const char *bulk2type(struct ptlrpc_request *req) { if (req->rq_bulk_read) @@ -3305,23 +3338,19 @@ static inline const char *bulk2type(struct ptlrpc_request *req) return "UNKNOWN"; } -int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, - struct l_wait_info *lwi) +int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc) { struct ptlrpc_request *req = desc->bd_req; - time64_t start = ktime_get_real_seconds(); + time64_t start = ktime_get_seconds(); time64_t deadline; int rc = 0; ENTRY; /* If there is eviction in progress, wait for it to finish. */ - if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) { - *lwi = LWI_INTR(NULL, NULL); - rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq, - !atomic_read(&exp->exp_obd->obd_evict_inprogress), - lwi); - } + wait_event_idle( + exp->exp_obd->obd_evict_inprogress_waitq, + !atomic_read(&exp->exp_obd->obd_evict_inprogress)); /* Check if client was evicted or reconnected already. */ if (exp->exp_failed || @@ -3341,7 +3370,7 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, } if (rc < 0) { - DEBUG_REQ(D_ERROR, req, "bulk %s failed: rc %d", + DEBUG_REQ(D_ERROR, req, "bulk %s failed: rc = %d", bulk2type(req), rc); RETURN(rc); } @@ -3357,28 +3386,27 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, deadline = req->rq_deadline; do { - time64_t timeoutl = deadline - ktime_get_real_seconds(); - long timeout_jiffies = timeoutl <= 0 ? - 1 : cfs_time_seconds(timeoutl); + time64_t timeoutl = deadline - ktime_get_seconds(); time64_t rq_deadline; - *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) || - exp->exp_failed || - exp->exp_conn_cnt > - lustre_msg_get_conn_cnt(req->rq_reqmsg), - lwi); - LASSERT(rc == 0 || rc == -ETIMEDOUT); + while (timeoutl >= 0 && + wait_event_idle_timeout( + desc->bd_waitq, + !ptlrpc_server_bulk_active(desc) || + exp->exp_failed || + exp->exp_conn_cnt > + lustre_msg_get_conn_cnt(req->rq_reqmsg), + timeoutl ? cfs_time_seconds(1) : 1) == 0) + timeoutl -= 1; + rc = timeoutl < 0 ? -ETIMEDOUT : 0; + /* Wait again if we changed rq_deadline. */ rq_deadline = READ_ONCE(req->rq_deadline); deadline = start + bulk_timeout; if (deadline > rq_deadline) deadline = rq_deadline; } while (rc == -ETIMEDOUT && - deadline > ktime_get_real_seconds()); + deadline > ktime_get_seconds()); if (rc == -ETIMEDOUT) { DEBUG_REQ(D_ERROR, req, "timeout on bulk %s after %lld%+llds",