X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lib.c;h=5c122fbbcf689ebbbc818a7792d7bff6151e2887;hp=3726f183077c539859eb571b3203755b9b8ef49d;hb=c1d465de13ccf0eda8020c88661c3cc4d78538ca;hpb=39a395681cf7da68474049babe9411b0325f46cf diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c index 3726f18..5c122fb 100644 --- a/lustre/ldlm/ldlm_lib.c +++ b/lustre/ldlm/ldlm_lib.c @@ -23,7 +23,7 @@ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2010, 2015, Intel Corporation. + * Copyright (c) 2010, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -54,71 +54,75 @@ static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid, int priority, int create) { - struct ptlrpc_connection *ptlrpc_conn; - struct obd_import_conn *imp_conn = NULL, *item; - int rc = 0; - ENTRY; + struct ptlrpc_connection *ptlrpc_conn; + struct obd_import_conn *imp_conn = NULL, *item; + lnet_nid_t nid4refnet = LNET_NID_ANY; + int rc = 0; + ENTRY; - if (!create && !priority) { - CDEBUG(D_HA, "Nothing to do\n"); - RETURN(-EINVAL); - } + if (!create && !priority) { + CDEBUG(D_HA, "Nothing to do\n"); + RETURN(-EINVAL); + } - ptlrpc_conn = ptlrpc_uuid_to_connection(uuid); - if (!ptlrpc_conn) { - CDEBUG(D_HA, "can't find connection %s\n", uuid->uuid); - RETURN (-ENOENT); - } + if (imp->imp_connection && + imp->imp_connection->c_remote_uuid.uuid[0] == 0) + /* nid4refnet is used to restrict network connections */ + nid4refnet = imp->imp_connection->c_self; + ptlrpc_conn = ptlrpc_uuid_to_connection(uuid, nid4refnet); + if (!ptlrpc_conn) { + CDEBUG(D_HA, "can't find connection %s\n", uuid->uuid); + RETURN(-ENOENT); + } - if (create) { - OBD_ALLOC(imp_conn, sizeof(*imp_conn)); - if (!imp_conn) { - GOTO(out_put, rc = -ENOMEM); - } - } + if (create) { + OBD_ALLOC(imp_conn, sizeof(*imp_conn)); + if (!imp_conn) + GOTO(out_put, rc = -ENOMEM); + } spin_lock(&imp->imp_lock); list_for_each_entry(item, &imp->imp_conn_list, oic_item) { - if (obd_uuid_equals(uuid, &item->oic_uuid)) { - if (priority) { + if (obd_uuid_equals(uuid, &item->oic_uuid)) { + if (priority) { list_del(&item->oic_item); list_add(&item->oic_item, - &imp->imp_conn_list); - item->oic_last_attempt = 0; - } - CDEBUG(D_HA, "imp %p@%s: found existing conn %s%s\n", - imp, imp->imp_obd->obd_name, uuid->uuid, - (priority ? ", moved to head" : "")); + &imp->imp_conn_list); + item->oic_last_attempt = 0; + } + CDEBUG(D_HA, "imp %p@%s: found existing conn %s%s\n", + imp, imp->imp_obd->obd_name, uuid->uuid, + (priority ? ", moved to head" : "")); spin_unlock(&imp->imp_lock); - GOTO(out_free, rc = 0); - } - } + GOTO(out_free, rc = 0); + } + } /* No existing import connection found for \a uuid. */ - if (create) { - imp_conn->oic_conn = ptlrpc_conn; - imp_conn->oic_uuid = *uuid; - imp_conn->oic_last_attempt = 0; - if (priority) + if (create) { + imp_conn->oic_conn = ptlrpc_conn; + imp_conn->oic_uuid = *uuid; + imp_conn->oic_last_attempt = 0; + if (priority) list_add(&imp_conn->oic_item, &imp->imp_conn_list); - else + else list_add_tail(&imp_conn->oic_item, - &imp->imp_conn_list); - CDEBUG(D_HA, "imp %p@%s: add connection %s at %s\n", - imp, imp->imp_obd->obd_name, uuid->uuid, - (priority ? "head" : "tail")); - } else { + &imp->imp_conn_list); + CDEBUG(D_HA, "imp %p@%s: add connection %s at %s\n", + imp, imp->imp_obd->obd_name, uuid->uuid, + (priority ? "head" : "tail")); + } else { spin_unlock(&imp->imp_lock); GOTO(out_free, rc = -ENOENT); } spin_unlock(&imp->imp_lock); - RETURN(0); + RETURN(0); out_free: - if (imp_conn) - OBD_FREE(imp_conn, sizeof(*imp_conn)); + if (imp_conn) + OBD_FREE(imp_conn, sizeof(*imp_conn)); out_put: - ptlrpc_connection_put(ptlrpc_conn); - RETURN(rc); + ptlrpc_connection_put(ptlrpc_conn); + RETURN(rc); } int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid) @@ -256,6 +260,7 @@ static int osc_on_mdt(char *obdname) * 1 - client UUID * 2 - server UUID * 3 - inactive-on-startup + * 4 - restrictive net */ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) { @@ -266,6 +271,8 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) char *name = obddev->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, + .c_remote_uuid.uuid[0] = 0 }; int rc; ENTRY; @@ -343,10 +350,12 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) init_rwsem(&cli->cl_sem); mutex_init(&cli->cl_mgc_mutex); - cli->cl_conn_count = 0; - memcpy(server_uuid.uuid, lustre_cfg_buf(lcfg, 2), - min_t(unsigned int, LUSTRE_CFG_BUFLEN(lcfg, 2), - sizeof(server_uuid))); + cli->cl_seq = NULL; + init_rwsem(&cli->cl_seq_rwsem); + cli->cl_conn_count = 0; + memcpy(server_uuid.uuid, lustre_cfg_buf(lcfg, 2), + min_t(unsigned int, LUSTRE_CFG_BUFLEN(lcfg, 2), + sizeof(server_uuid))); cli->cl_dirty_pages = 0; cli->cl_avail_grant = 0; @@ -382,6 +391,9 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) 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 @@ -400,6 +412,8 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) * 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; @@ -417,7 +431,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) 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); @@ -427,6 +441,8 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) init_waitqueue_head(&cli->cl_mod_rpcs_waitq); cli->cl_mod_tag_bitmap = NULL; + INIT_LIST_HEAD(&cli->cl_chg_dev_linkage); + if (connect_op == MDS_CONNECT) { cli->cl_max_mod_rpcs_in_flight = cli->cl_max_rpcs_in_flight - 1; OBD_ALLOC(cli->cl_mod_tag_bitmap, @@ -453,11 +469,26 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) LUSTRE_CFG_BUFLEN(lcfg, 1)); class_import_put(imp); - rc = client_import_add_conn(imp, &server_uuid, 1); - if (rc) { - CERROR("can't add initial connection\n"); - GOTO(err_import, rc); - } + if (lustre_cfg_buf(lcfg, 4)) { + __u32 refnet = libcfs_str2net(lustre_cfg_string(lcfg, 4)); + + 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), + rc); + GOTO(err_import, rc); + } + fake_conn.c_self = LNET_MKNID(refnet, 0); + imp->imp_connection = &fake_conn; + } + + rc = client_import_add_conn(imp, &server_uuid, 1); + if (rc) { + CERROR("can't add initial connection\n"); + GOTO(err_import, rc); + } + imp->imp_connection = NULL; cli->cl_import = imp; /* cli->cl_max_mds_easize updated by mdc_init_ea_size() */ @@ -533,7 +564,6 @@ int client_connect_import(const struct lu_env *env, struct obd_connect_data *ocd; struct lustre_handle conn = { 0 }; int rc; - bool is_mdc = false; ENTRY; *exp = NULL; @@ -558,18 +588,12 @@ int client_connect_import(const struct lu_env *env, ocd = &imp->imp_connect_data; if (data) { *ocd = *data; - is_mdc = strncmp(imp->imp_obd->obd_type->typ_name, - LUSTRE_MDC_NAME, 3) == 0; - if (is_mdc) - data->ocd_connect_flags |= OBD_CONNECT_MULTIMODRPCS; imp->imp_connect_flags_orig = data->ocd_connect_flags; imp->imp_connect_flags2_orig = data->ocd_connect_flags2; } rc = ptlrpc_connect_import(imp); if (rc != 0) { - if (data && is_mdc) - data->ocd_connect_flags &= ~OBD_CONNECT_MULTIMODRPCS; LASSERT(imp->imp_state == LUSTRE_IMP_DISCON); GOTO(out_ldlm, rc); } @@ -580,10 +604,7 @@ int client_connect_import(const struct lu_env *env, 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; - /* clear the flag as it was not set and is not known - * by upper layers */ - if (is_mdc) - data->ocd_connect_flags &= ~OBD_CONNECT_MULTIMODRPCS; + data->ocd_connect_flags2 = ocd->ocd_connect_flags2; } ptlrpc_pinger_add_import(imp); @@ -701,7 +722,10 @@ int server_disconnect_export(struct obd_export *exp) spin_lock(&svcpt->scp_rep_lock); list_del_init(&rs->rs_exp_list); + spin_lock(&rs->rs_lock); + /* clear rs_convert_lock to make sure rs is handled and put */ + rs->rs_convert_lock = 0; ptlrpc_schedule_difficult_reply(rs); spin_unlock(&rs->rs_lock); @@ -723,12 +747,12 @@ static int target_handle_reconnect(struct lustre_handle *conn, { 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; @@ -762,16 +786,16 @@ static int target_handle_reconnect(struct lustre_handle *conn, GOTO(out_already, rc); } - now = cfs_time_current(); - deadline = cfs_timer_deadline(&target->obd_recovery_timer); - if (cfs_time_before(now, deadline)) { - struct target_distribute_txn_data *tdtd = - class_exp2tgt(exp)->lut_tdtd; + now = ktime_get_seconds(); + deadline = cfs_duration_sec(target->obd_recovery_timer.expires); + 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, @@ -779,7 +803,7 @@ static int target_handle_reconnect(struct lustre_handle *conn, 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", @@ -788,7 +812,7 @@ static int target_handle_reconnect(struct lustre_handle *conn, 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); @@ -796,9 +820,9 @@ static int target_handle_reconnect(struct lustre_handle *conn, 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); @@ -812,20 +836,6 @@ out_already: RETURN(EALREADY); } -void target_client_add_cb(struct obd_device *obd, __u64 transno, void *cb_data, - int error) -{ - struct obd_export *exp = cb_data; - - CDEBUG(D_RPCTRACE, "%s: committing for initial connect of %s\n", - obd->obd_name, exp->exp_client_uuid.uuid); - - spin_lock(&exp->exp_lock); - exp->exp_need_sync = 0; - spin_unlock(&exp->exp_lock); - class_export_cb_put(exp); -} - static void check_and_start_recovery_timer(struct obd_device *obd, struct ptlrpc_request *req, int new_client); @@ -946,7 +956,6 @@ int target_handle_connect(struct ptlrpc_request *req) * reconnect case */ struct lustre_handle conn; struct lustre_handle *tmp; - struct obd_uuid tgtuuid; struct obd_uuid cluuid; char *str; int rc = 0; @@ -955,7 +964,6 @@ int target_handle_connect(struct ptlrpc_request *req) 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; @@ -969,11 +977,7 @@ int target_handle_connect(struct ptlrpc_request *req) 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 " @@ -985,6 +989,9 @@ int target_handle_connect(struct ptlrpc_request *req) } spin_lock(&target->obd_dev_lock); + + target->obd_conn_inprogress++; + if (target->obd_stopping || !target->obd_set_up) { spin_unlock(&target->obd_dev_lock); @@ -1006,13 +1013,6 @@ int target_handle_connect(struct ptlrpc_request *req) 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); @@ -1051,32 +1051,17 @@ int target_handle_connect(struct ptlrpc_request *req) */ 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 @@ -1094,7 +1079,8 @@ int target_handle_connect(struct ptlrpc_request *req) * * 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) && @@ -1228,11 +1214,11 @@ no_export: 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. */ @@ -1258,27 +1244,26 @@ no_export: /* 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 = cfs_timer_deadline(&target->obd_recovery_timer); - t = cfs_time_sub(t, cfs_time_current()); - t = cfs_duration_sec(t); + t = target->obd_recovery_timer.expires; + t = cfs_duration_sec(target->obd_recovery_timer.expires); + 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: @@ -1333,37 +1318,26 @@ 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)); + 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); - - /* 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, @@ -1437,12 +1411,11 @@ out: 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); @@ -1571,18 +1544,20 @@ static void target_finish_recovery(struct lu_target *lut) /* 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 " + time64_t now = ktime_get_real_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, - (int)elapsed_time / 60, (int)elapsed_time % 60, + (s64)elapsed_time / 60, (s64)elapsed_time % 60, obd->obd_max_recoverable_clients, atomic_read(&obd->obd_connected_clients), obd->obd_stale_clients, obd->obd_stale_clients == 1 ? "was" : "were"); } - ldlm_reprocess_all_ns(obd->obd_namespace); + ldlm_reprocess_recovery_done(obd->obd_namespace); spin_lock(&obd->obd_recovery_task_lock); if (!list_empty(&obd->obd_req_replay_queue) || !list_empty(&obd->obd_lock_replay_queue) || @@ -1599,15 +1574,16 @@ static void target_finish_recovery(struct lu_target *lut) } spin_unlock(&obd->obd_recovery_task_lock); - obd->obd_recovery_end = cfs_time_current_sec(); + 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; } @@ -1705,7 +1681,7 @@ EXPORT_SYMBOL(target_cleanup_recovery); 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); + del_timer(&obd->obd_recovery_timer); } static void target_start_recovery_timer(struct obd_device *obd) @@ -1726,19 +1702,18 @@ static void target_start_recovery_timer(struct obd_device *obd) return; } - cfs_timer_arm(&obd->obd_recovery_timer, - cfs_time_shift(obd->obd_recovery_timeout)); - obd->obd_recovery_start = cfs_time_current_sec(); + mod_timer(&obd->obd_recovery_timer, + jiffies + cfs_time_seconds(obd->obd_recovery_timeout)); + obd->obd_recovery_start = ktime_get_real_seconds(); spin_unlock(&obd->obd_dev_lock); - LCONSOLE_WARN("%s: Will be in recovery for at least %d:%.02d, " - "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": ""); + LCONSOLE_WARN("%s: Will be in recovery for at least %llu:%02llu, 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": ""); } /** @@ -1747,24 +1722,25 @@ static void target_start_recovery_timer(struct obd_device *obd) * 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, bool extend) +static void extend_recovery_timer(struct obd_device *obd, time64_t drt, + bool extend) { - cfs_time_t now; - cfs_time_t end; - cfs_duration_t left; - int to; + time64_t now; + time64_t end; + time64_t left; + time64_t to; spin_lock(&obd->obd_dev_lock); if (!obd->obd_recovering || obd->obd_abort_recovery) { spin_unlock(&obd->obd_dev_lock); - return; - } - LASSERT(obd->obd_recovery_start != 0); + return; + } + LASSERT(obd->obd_recovery_start != 0); - now = cfs_time_current_sec(); - to = obd->obd_recovery_timeout; - end = obd->obd_recovery_start + to; - left = cfs_time_sub(end, now); + 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; @@ -1772,18 +1748,22 @@ static void extend_recovery_timer(struct obd_device *obd, int drt, bool extend) to = drt; } - if (to > obd->obd_recovery_time_hard) - to = obd->obd_recovery_time_hard; + 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", + obd->obd_name, to, extend); + } + if (obd->obd_recovery_timeout < to) { obd->obd_recovery_timeout = to; end = obd->obd_recovery_start + to; - cfs_timer_arm(&obd->obd_recovery_timer, - cfs_time_shift(end - now)); + mod_timer(&obd->obd_recovery_timer, + jiffies + cfs_time_seconds(end - now)); } spin_unlock(&obd->obd_dev_lock); - CDEBUG(D_HA, "%s: recovery timer will expire in %u seconds\n", - obd->obd_name, (unsigned)cfs_time_sub(end, now)); + CDEBUG(D_HA, "%s: recovery timer will expire in %lld seconds\n", + obd->obd_name, (s64)(end - now)); } /* Reset the timer with each new client connection */ @@ -1802,7 +1782,7 @@ check_and_start_recovery_timer(struct obd_device *obd, 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) @@ -1814,9 +1794,14 @@ check_and_start_recovery_timer(struct obd_device *obd, 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) && + service_time < at_extra) + service_time = at_extra; + /* We expect other clients to timeout within service_time, then try * to reconnect, then try the failover server. The max delay between * connect attempts is SWITCH_MAX + SWITCH_INC + INITIAL. */ @@ -1982,9 +1967,20 @@ static int target_recovery_overseer(struct lu_target *lut, { struct obd_device *obd = lut->lut_obd; struct target_distribute_txn_data *tdtd; + time64_t last = 0; + time64_t now; repeat: - if ((obd->obd_recovery_start != 0) && (cfs_time_current_sec() >= - (obd->obd_recovery_start + obd->obd_recovery_time_hard))) { + 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; /* Only abort the recovery if there are no update recovery @@ -2153,7 +2149,7 @@ static void handle_recovery_req(struct ptlrpc_thread *thread, /* 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 @@ -2207,9 +2203,8 @@ static int check_for_recovery_ready(struct lu_target *lut) * timer expired, and some clients got evicted */ extend_recovery_timer(obd, obd->obd_recovery_timeout, true); - CDEBUG(D_HA, "%s update recovery is not ready," - " extend recovery %d\n", obd->obd_name, - obd->obd_recovery_timeout); + CDEBUG(D_HA, "%s update recovery is not ready, extend recovery %llu\n", + obd->obd_name, obd->obd_recovery_timeout); return 0; } } @@ -2612,9 +2607,9 @@ static void target_recovery_expired(unsigned long castmeharder) { struct obd_device *obd = (struct obd_device *)castmeharder; CDEBUG(D_HA, "%s: recovery timed out; %d clients are still in recovery" - " after %lds (%d clients connected)\n", + " after %llus (%d clients connected)\n", obd->obd_name, atomic_read(&obd->obd_lock_replay_clients), - cfs_time_current_sec()- obd->obd_recovery_start, + (s64)(ktime_get_real_seconds() - obd->obd_recovery_start), atomic_read(&obd->obd_connected_clients)); obd->obd_recovery_expired = 1; @@ -2623,13 +2618,16 @@ static void target_recovery_expired(unsigned long castmeharder) void target_recovery_init(struct lu_target *lut, svc_handler_t handler) { - struct obd_device *obd = lut->lut_obd; + struct obd_device *obd = lut->lut_obd; - if (obd->obd_max_recoverable_clients == 0) { - /** Update server last boot epoch */ - tgt_boot_epoch_update(lut); - return; - } + if (lut->lut_bottom->dd_rdonly) + return; + + if (obd->obd_max_recoverable_clients == 0) { + /** Update server last boot epoch */ + tgt_boot_epoch_update(lut); + return; + } CDEBUG(D_HA, "RECOVERY: service %s, %d recoverable clients, " "last_transno %llu\n", obd->obd_name, @@ -2639,8 +2637,9 @@ void target_recovery_init(struct lu_target *lut, svc_handler_t handler) obd->obd_recovery_start = 0; obd->obd_recovery_end = 0; - cfs_timer_init(&obd->obd_recovery_timer, target_recovery_expired, obd); - target_start_recovery_thread(lut, handler); + setup_timer(&obd->obd_recovery_timer, target_recovery_expired, + (unsigned long)obd); + target_start_recovery_thread(lut, handler); } EXPORT_SYMBOL(target_recovery_init); @@ -2695,6 +2694,17 @@ int target_queue_recovery_request(struct ptlrpc_request *req, 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); @@ -3142,10 +3152,10 @@ static inline const char *bulk2type(struct ptlrpc_request *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; @@ -3192,12 +3202,13 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, 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) || @@ -3211,13 +3222,13 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, 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",