X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lib.c;h=b4e9d61f8fd7ec1f11fe76d7d875bf1195b8aaa6;hb=0a713200f663ba86c2a7f5bc65726cd8884fc79f;hp=3b2ab47eb92a0311e7875c2f4020325160174bf6;hpb=10d49b6e200bb4234a2611e2882e646bf811cd68;p=fs%2Flustre-release.git diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c index 3b2ab47..b4e9d61 100644 --- a/lustre/ldlm/ldlm_lib.c +++ b/lustre/ldlm/ldlm_lib.c @@ -1,25 +1,37 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Copyright (c) 2003 Cluster File Systems, Inc. + * GPL HEADER START * - * This file is part of the Lustre file system, http://www.lustre.org - * Lustre is a trademark of Cluster File Systems, Inc. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * You may have signed or agreed to another license before downloading - * this software. If so, you are bound by the terms and conditions - * of that agreement, and the following does not apply to you. See the - * LICENSE file included with this distribution for more information. + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. * - * If you did not agree to a different license, then this copy of Lustre - * is open source software; you can redistribute it and/or modify it - * under the terms of version 2 of the GNU General Public License as - * published by the Free Software Foundation. + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * In either case, Lustre is distributed in the hope that it will be - * useful, but WITHOUT ANY WARRANTY; without even the implied warranty - * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * license text for more details. + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + * GPL HEADER END + */ +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Use is subject to license terms. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. */ #ifndef EXPORT_SYMTAB @@ -39,7 +51,6 @@ #include #include "ldlm_internal.h" - /* @priority: if non-zero, move the selected to the list head * @create: if zero, only search in existed connections */ @@ -88,7 +99,7 @@ static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid, if (create) { imp_conn->oic_conn = ptlrpc_conn; imp_conn->oic_uuid = *uuid; - item->oic_last_attempt = 0; + imp_conn->oic_last_attempt = 0; if (priority) list_add(&imp_conn->oic_item, &imp->imp_conn_list); else @@ -107,7 +118,7 @@ out_free: if (imp_conn) OBD_FREE(imp_conn, sizeof(*imp_conn)); out_put: - ptlrpc_put_connection(ptlrpc_conn); + ptlrpc_connection_put(ptlrpc_conn); RETURN(rc); } @@ -150,20 +161,20 @@ int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid) GOTO(out, rc = -EBUSY); } - ptlrpc_put_connection(imp->imp_connection); + ptlrpc_connection_put(imp->imp_connection); imp->imp_connection = NULL; dlmexp = class_conn2export(&imp->imp_dlm_handle); if (dlmexp && dlmexp->exp_connection) { LASSERT(dlmexp->exp_connection == imp_conn->oic_conn); - ptlrpc_put_connection(dlmexp->exp_connection); + ptlrpc_connection_put(dlmexp->exp_connection); dlmexp->exp_connection = NULL; } } list_del(&imp_conn->oic_item); - ptlrpc_put_connection(imp_conn->oic_conn); + ptlrpc_connection_put(imp_conn->oic_conn); OBD_FREE(imp_conn, sizeof(*imp_conn)); CDEBUG(D_HA, "imp %p@%s: remove connection %s\n", imp, imp->imp_obd->obd_name, uuid->uuid); @@ -177,7 +188,7 @@ out: RETURN(rc); } -static void destroy_import(struct obd_import *imp) +void client_destroy_import(struct obd_import *imp) { /* drop security policy instance after all rpc finished/aborted * to let all busy contexts be released. */ @@ -210,14 +221,21 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) rq_portal = OST_REQUEST_PORTAL; rp_portal = OSC_REPLY_PORTAL; connect_op = OST_CONNECT; + cli->cl_sp_me = LUSTRE_SP_CLI; + cli->cl_sp_to = LUSTRE_SP_OST; } else if (!strcmp(name, LUSTRE_MDC_NAME)) { rq_portal = MDS_REQUEST_PORTAL; rp_portal = MDC_REPLY_PORTAL; connect_op = MDS_CONNECT; + cli->cl_sp_me = LUSTRE_SP_CLI; + cli->cl_sp_to = LUSTRE_SP_MDT; } else if (!strcmp(name, LUSTRE_MGC_NAME)) { rq_portal = MGS_REQUEST_PORTAL; rp_portal = MGC_REPLY_PORTAL; connect_op = MGS_CONNECT; + cli->cl_sp_me = LUSTRE_SP_MGC; + cli->cl_sp_to = LUSTRE_SP_MGS; + cli->cl_flvr_mgc.sf_rpc = SPTLRPC_FLVR_INVALID; } else { CERROR("unknown client OBD type \"%s\", can't setup\n", name); @@ -244,10 +262,8 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) RETURN(-EINVAL); } - sema_init(&cli->cl_sem, 1); + init_rwsem(&cli->cl_sem); sema_init(&cli->cl_mgc_sem, 1); - sptlrpc_rule_set_init(&cli->cl_sptlrpc_rset); - cli->cl_sec_part = LUSTRE_SP_ANY; cli->cl_conn_count = 0; memcpy(server_uuid.uuid, lustre_cfg_buf(lcfg, 2), min_t(unsigned int, LUSTRE_CFG_BUFLEN(lcfg, 2), @@ -261,6 +277,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) cli->cl_dirty_max = num_physpages << (CFS_PAGE_SHIFT - 3); CFS_INIT_LIST_HEAD(&cli->cl_cache_waiters); CFS_INIT_LIST_HEAD(&cli->cl_loi_ready_list); + CFS_INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list); CFS_INIT_LIST_HEAD(&cli->cl_loi_write_list); CFS_INIT_LIST_HEAD(&cli->cl_loi_read_list); client_obd_list_lock_init(&cli->cl_loi_list_lock); @@ -333,7 +350,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) cli->cl_import = imp; /* cli->cl_max_mds_{easize,cookiesize} updated by mdc_init_ea_size() */ - cli->cl_max_mds_easize = sizeof(struct lov_mds_md); + cli->cl_max_mds_easize = sizeof(struct lov_mds_md_v3); cli->cl_max_mds_cookiesize = sizeof(struct llog_cookie); if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) { @@ -354,7 +371,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) err_import: class_destroy_import(imp); err_ldlm: - ldlm_put_ref(0); + ldlm_put_ref(); err: RETURN(rc); @@ -363,44 +380,51 @@ err: int client_obd_cleanup(struct obd_device *obddev) { ENTRY; - sptlrpc_rule_set_free(&obddev->u.cli.cl_sptlrpc_rset); - ldlm_put_ref(obddev->obd_force); + ldlm_put_ref(); RETURN(0); } /* ->o_connect() method for client side (OSC and MDC and MGC) */ int client_connect_import(const struct lu_env *env, - struct lustre_handle *dlm_handle, + struct obd_export **exp, struct obd_device *obd, struct obd_uuid *cluuid, - struct obd_connect_data *data) + struct obd_connect_data *data, void *localdata) { struct client_obd *cli = &obd->u.cli; struct obd_import *imp = cli->cl_import; - struct obd_export *exp; struct obd_connect_data *ocd; struct ldlm_namespace *to_be_freed = NULL; + struct lustre_handle conn = { 0 }; int rc; ENTRY; - mutex_down(&cli->cl_sem); - rc = class_connect(dlm_handle, obd, cluuid); + *exp = NULL; + down_write(&cli->cl_sem); + if (cli->cl_conn_count > 0 ) + GOTO(out_sem, rc = -EALREADY); + + rc = class_connect(&conn, obd, cluuid); if (rc) GOTO(out_sem, rc); - + cli->cl_conn_count++; - if (cli->cl_conn_count > 1) - GOTO(out_sem, rc); - exp = class_conn2export(dlm_handle); + *exp = class_conn2export(&conn); if (obd->obd_namespace != NULL) CERROR("already have namespace!\n"); - obd->obd_namespace = ldlm_namespace_new(obd->obd_name, + + /* + * Deadlock case - bug 18380 + */ + up_write(&cli->cl_sem); + obd->obd_namespace = ldlm_namespace_new(obd, obd->obd_name, LDLM_NAMESPACE_CLIENT, LDLM_NAMESPACE_GREEDY); + down_write(&cli->cl_sem); if (obd->obd_namespace == NULL) GOTO(out_disco, rc = -ENOMEM); - imp->imp_dlm_handle = *dlm_handle; + imp->imp_dlm_handle = conn; rc = ptlrpc_init_import(imp); if (rc != 0) GOTO(out_ldlm, rc); @@ -416,7 +440,7 @@ int client_connect_import(const struct lu_env *env, LASSERT (imp->imp_state == LUSTRE_IMP_DISCON); GOTO(out_ldlm, rc); } - LASSERT(exp->exp_connection); + LASSERT((*exp)->exp_connection); if (data) { LASSERTF((ocd->ocd_connect_flags & data->ocd_connect_flags) == @@ -431,19 +455,18 @@ int client_connect_import(const struct lu_env *env, if (rc) { out_ldlm: - ldlm_namespace_free_prior(obd->obd_namespace); + ldlm_namespace_free_prior(obd->obd_namespace, imp, 0); to_be_freed = obd->obd_namespace; obd->obd_namespace = NULL; out_disco: cli->cl_conn_count--; - class_disconnect(exp); - } else { - class_export_put(exp); + class_disconnect(*exp); + *exp = NULL; } out_sem: - mutex_up(&cli->cl_sem); + up_write(&cli->cl_sem); if (to_be_freed) - ldlm_namespace_free_post(to_be_freed, 1); + ldlm_namespace_free_post(to_be_freed); return rc; } @@ -466,16 +489,16 @@ int client_disconnect_export(struct obd_export *exp) cli = &obd->u.cli; imp = cli->cl_import; - mutex_down(&cli->cl_sem); + down_write(&cli->cl_sem); if (!cli->cl_conn_count) { CERROR("disconnecting disconnected device (%s)\n", obd->obd_name); - GOTO(out_sem, rc = -EINVAL); + GOTO(out_disconnect, rc = -EINVAL); } cli->cl_conn_count--; if (cli->cl_conn_count) - GOTO(out_no_disconnect, rc = 0); + GOTO(out_disconnect, rc = 0); /* Mark import deactivated now, so we don't try to reconnect if any * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't @@ -494,11 +517,17 @@ int client_disconnect_export(struct obd_export *exp) ldlm_cli_cancel_unused(obd->obd_namespace, NULL, obd->obd_force ? LDLM_FL_LOCAL_ONLY:0, NULL); - ldlm_namespace_free_prior(obd->obd_namespace); + ldlm_namespace_free_prior(obd->obd_namespace, imp, obd->obd_force); to_be_freed = obd->obd_namespace; } + /* + * there's no necessary to hold sem during diconnecting an import, + * and actually it may cause deadlock in gss. + */ + up_write(&cli->cl_sem); rc = ptlrpc_disconnect_import(imp, 0); + down_write(&cli->cl_sem); ptlrpc_invalidate_import(imp); /* set obd_namespace to NULL only after invalidate, because we can have @@ -506,19 +535,25 @@ int client_disconnect_export(struct obd_export *exp) * in obd_namespace. bug 14260 */ obd->obd_namespace = NULL; - ptlrpc_free_rq_pool(imp->imp_rq_pool); - destroy_import(imp); + if (imp->imp_rq_pool) { + ptlrpc_free_rq_pool(imp->imp_rq_pool); + imp->imp_rq_pool = NULL; + } + client_destroy_import(imp); cli->cl_import = NULL; EXIT; - out_no_disconnect: + + out_disconnect: + /* use server style - class_disconnect should be always called for + * o_disconnect */ err = class_disconnect(exp); if (!rc && err) rc = err; - out_sem: - mutex_up(&cli->cl_sem); + + up_write(&cli->cl_sem); if (to_be_freed) - ldlm_namespace_free_post(to_be_freed, obd->obd_force); + ldlm_namespace_free_post(to_be_freed); RETURN(rc); } @@ -527,11 +562,12 @@ int client_disconnect_export(struct obd_export *exp) * from old lib/target.c * -------------------------------------------------------------------------- */ -int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp, - struct obd_uuid *cluuid, int initial_conn) +static int target_handle_reconnect(struct lustre_handle *conn, + struct obd_export *exp, + struct obd_uuid *cluuid) { ENTRY; - if (exp->exp_connection && exp->exp_imp_reverse && !initial_conn) { + if (exp->exp_connection && exp->exp_imp_reverse) { struct lustre_handle *hdl; hdl = &exp->exp_imp_reverse->imp_remote_handle; /* Might be a re-connect after a partition. */ @@ -576,6 +612,10 @@ void target_client_add_cb(struct obd_device *obd, __u64 transno, void *cb_data, spin_unlock(&exp->exp_lock); } EXPORT_SYMBOL(target_client_add_cb); +static void +target_start_and_reset_recovery_timer(struct obd_device *obd, + struct ptlrpc_request *req, + int new_client); int target_handle_connect(struct ptlrpc_request *req) { @@ -589,8 +629,9 @@ int target_handle_connect(struct ptlrpc_request *req) struct obd_uuid remote_uuid; char *str; int rc = 0; - int initial_conn = 0; + int mds_conn = 0; struct obd_connect_data *data, *tmpdata; + lnet_nid_t *client_nid = NULL; ENTRY; OBD_RACE(OBD_FAIL_TGT_CONN_RACE); @@ -625,7 +666,7 @@ int target_handle_connect(struct ptlrpc_request *req) /* 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. */ - targref = class_incref(target); + targref = class_incref(target, __FUNCTION__, cfs_current()); str = req_capsule_client_get(&req->rq_pill, &RMF_CLUUID); @@ -694,48 +735,66 @@ int target_handle_connect(struct ptlrpc_request *req) } } - if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL) - initial_conn = 1; + if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL) && + (data->ocd_connect_flags & OBD_CONNECT_MDS)) + mds_conn = 1; /* lctl gets a backstage, all-access pass. */ if (obd_uuid_equals(&cluuid, &target->obd_uuid)) goto dont_check_exports; - spin_lock(&target->obd_dev_lock); - export = lustre_hash_get_object_by_key(target->obd_uuid_hash_body, &cluuid); + export = lustre_hash_lookup(target->obd_uuid_hash, &cluuid); + if (!export) + goto no_export; - if (export != NULL && export->exp_connecting) { /* bug 9635, et. al. */ + /* we've found an export in the hash */ + if (export->exp_connecting) { /* bug 9635, et. al. */ CWARN("%s: exp %p already connecting\n", export->exp_obd->obd_name, export); class_export_put(export); export = NULL; rc = -EALREADY; - } else if (export != NULL && export->exp_connection != NULL && + } else if (mds_conn && export->exp_connection) { + if (req->rq_peer.nid != export->exp_connection->c_peer.nid) + /* mds reconnected after failover */ + CWARN("%s: received MDS connection from NID %s," + " removing former export from NID %s\n", + target->obd_name, libcfs_nid2str(req->rq_peer.nid), + libcfs_nid2str(export->exp_connection->c_peer.nid)); + else + /* new mds connection from the same nid */ + CWARN("%s: received new MDS connection from NID %s," + " removing former export from same NID\n", + target->obd_name, libcfs_nid2str(req->rq_peer.nid)); + class_fail_export(export); + class_export_put(export); + export = NULL; + rc = 0; + } else if (export->exp_connection != NULL && req->rq_peer.nid != export->exp_connection->c_peer.nid) { - /* make darn sure this is coming from the same peer - * if the UUIDs matched */ - CWARN("%s: cookie %s seen on new NID %s when " - "existing NID %s is already connected\n", - target->obd_name, cluuid.uuid, - libcfs_nid2str(req->rq_peer.nid), - libcfs_nid2str(export->exp_connection->c_peer.nid)); - class_export_put(export); - export = NULL; - rc = -EALREADY; - } else if (export != NULL) { + /* in mds failover we have static uuid but nid can be + * changed*/ + CWARN("%s: cookie %s seen on new NID %s when " + "existing NID %s is already connected\n", + target->obd_name, cluuid.uuid, + libcfs_nid2str(req->rq_peer.nid), + libcfs_nid2str(export->exp_connection->c_peer.nid)); + rc = -EALREADY; + class_export_put(export); + export = NULL; + } else { spin_lock(&export->exp_lock); export->exp_connecting = 1; spin_unlock(&export->exp_lock); class_export_put(export); - spin_unlock(&target->obd_dev_lock); LASSERT(export->exp_obd == target); - rc = target_handle_reconnect(&conn, export, &cluuid, initial_conn); + rc = target_handle_reconnect(&conn, export, &cluuid); } /* If we found an export, we already unlocked. */ if (!export) { - spin_unlock(&target->obd_dev_lock); +no_export: OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_CONNECT, 2 * obd_timeout); } else if (req->rq_export == NULL && atomic_read(&export->exp_rpc_count) > 0) { @@ -746,50 +805,60 @@ int target_handle_connect(struct ptlrpc_request *req) GOTO(out, rc = -EBUSY); } else if (req->rq_export != NULL && (atomic_read(&export->exp_rpc_count) > 1)) { + /* the current connect rpc has increased exp_rpc_count */ CWARN("%s: refuse reconnection from %s@%s to 0x%p/%d\n", target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), - export, atomic_read(&export->exp_rpc_count)); + export, atomic_read(&export->exp_rpc_count) - 1); + spin_lock(&export->exp_lock); + if (req->rq_export->exp_conn_cnt < + lustre_msg_get_conn_cnt(req->rq_reqmsg)) + /* try to abort active requests */ + req->rq_export->exp_abort_active_req = 1; + spin_unlock(&export->exp_lock); GOTO(out, rc = -EBUSY); - } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1 && - !initial_conn) { + } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1) { CERROR("%s: NID %s (%s) reconnected with 1 conn_cnt; " "cookies not random?\n", target->obd_name, libcfs_nid2str(req->rq_peer.nid), cluuid.uuid); GOTO(out, rc = -EALREADY); } else { OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_RECONNECT, 2 * obd_timeout); - if (req->rq_export == NULL && initial_conn) - export->exp_last_request_time = - max(export->exp_last_request_time, - (time_t)CURRENT_SECONDS); } - /* We want to handle EALREADY but *not* -EALREADY from - * target_handle_reconnect(), return reconnection state in a flag */ - if (rc == EALREADY) { - lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT); - rc = 0; - } else if (rc) { + if (rc < 0) { GOTO(out, rc); } - /* Tell the client if we're in recovery. */ - /* If this is the first client, start the recovery timer */ + CWARN("%s: connection from %s@%s %st"LPU64" 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)CURRENT_SECONDS, + export, (long)cfs_time_current_sec(), export ? (long)export->exp_last_request_time : 0); - + /* Tell the client if we're in recovery. */ if (target->obd_recovering) { lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING); - target_start_recovery_timer(target); + /* If this is the first time a client connects, + reset the recovery timer */ + if (rc == 0) + target_start_and_reset_recovery_timer(target, req, + !export); + } + + /* We want to handle EALREADY but *not* -EALREADY from + * target_handle_reconnect(), return reconnection state in a flag */ + if (rc == EALREADY) { + lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECONNECT); + rc = 0; + } else { + LASSERT(rc == 0); } /* Tell the client if we support replayable requests */ if (target->obd_replayable) lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_REPLAYABLE); + client_nid = &req->rq_peer.nid; if (export == NULL) { if (target->obd_recovering) { @@ -798,7 +867,7 @@ int target_handle_connect(struct ptlrpc_request *req) t = cfs_timer_deadline(&target->obd_recovery_timer); t = cfs_time_sub(t, cfs_time_current()); CERROR("%s: denying connection for new client %s (%s): " - "%d clients in recovery for %lds\n", + "%d clients in recovery for "CFS_TIME_T"s\n", target->obd_name, libcfs_nid2str(req->rq_peer.nid), cluuid.uuid, target->obd_recoverable_clients, @@ -807,11 +876,17 @@ int target_handle_connect(struct ptlrpc_request *req) } else { dont_check_exports: rc = obd_connect(req->rq_svc_thread->t_env, - &conn, target, &cluuid, data); + &export, target, &cluuid, data, + client_nid); + if (rc == 0) + conn.cookie = export->exp_handle.h_cookie; } } else { rc = obd_reconnect(req->rq_svc_thread->t_env, - export, target, &cluuid, data); + export, target, &cluuid, data, client_nid); + if (rc == 0) + /* prevous done via class_conn2export */ + class_export_get(export); } if (rc) GOTO(out, rc); @@ -823,21 +898,12 @@ dont_check_exports: //data->ocd_connect_flags &= OBD_CONNECT_SUPPORTED; *tmpdata = *data; } - + /* If all else goes well, this is our RPC return code. */ req->rq_status = 0; lustre_msg_set_handle(req->rq_repmsg, &conn); - /* ownership of this export ref transfers to the request AFTER we - * drop any previous reference the request had, but we don't want - * that to go to zero before we get our new export reference. */ - export = class_conn2export(&conn); - if (!export) { - DEBUG_REQ(D_ERROR, req, "Missing export!"); - GOTO(out, rc = -ENODEV); - } - /* If the client and the server are the same node, we will already * have an export that really points to the client's DLM export, * because we have a shared handles table. @@ -851,9 +917,7 @@ dont_check_exports: req->rq_export = export; spin_lock(&export->exp_lock); - if (initial_conn) { - lustre_msg_set_conn_cnt(req->rq_repmsg, export->exp_conn_cnt + 1); - } else if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) { + if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) { spin_unlock(&export->exp_lock); CERROR("%s: %s already connected at higher conn_cnt: %d > %d\n", cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), @@ -863,6 +927,7 @@ dont_check_exports: GOTO(out, rc = -EALREADY); } export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg); + export->exp_abort_active_req = 0; /* request from liblustre? Don't evict it for not pinging. */ if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) { @@ -877,18 +942,15 @@ dont_check_exports: } if (export->exp_connection != NULL) - ptlrpc_put_connection(export->exp_connection); - export->exp_connection = ptlrpc_get_connection(req->rq_peer, + ptlrpc_connection_put(export->exp_connection); + export->exp_connection = ptlrpc_connection_get(req->rq_peer, req->rq_self, &remote_uuid); - - spin_lock(&target->obd_dev_lock); - /* Export might be hashed already, e.g. if this is reconnect */ - if (hlist_unhashed(&export->exp_nid_hash)) - lustre_hash_additem(export->exp_obd->obd_nid_hash_body, - &export->exp_connection->c_peer.nid, - &export->exp_nid_hash); - spin_unlock(&target->obd_dev_lock); + if (hlist_unhashed(&export->exp_nid_hash)) { + lustre_hash_add_unique(export->exp_obd->obd_nid_hash, + &export->exp_connection->c_peer.nid, + &export->exp_nid_hash); + } spin_lock_bh(&target->obd_processing_task_lock); if (target->obd_recovering && !export->exp_in_recovery) { @@ -911,7 +973,7 @@ dont_check_exports: atomic_inc(&target->obd_lock_replay_clients); if (target->obd_connected_clients == target->obd_max_recoverable_clients) - wake_up(&target->obd_next_transno_waitq); + cfs_waitq_signal(&target->obd_next_transno_waitq); } spin_unlock_bh(&target->obd_processing_task_lock); tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN); @@ -919,7 +981,7 @@ dont_check_exports: if (export->exp_imp_reverse != NULL) { /* destroyed import can be still referenced in ctxt */ - obd_set_info_async(export, strlen(KEY_REVIMP_UPD), + obd_set_info_async(export, sizeof(KEY_REVIMP_UPD), KEY_REVIMP_UPD, 0, NULL, NULL); /* in some recovery senarios, previous ctx init rpc handled @@ -935,7 +997,7 @@ dont_check_exports: */ sptlrpc_import_inval_all_ctx(export->exp_imp_reverse); - destroy_import(export->exp_imp_reverse); + client_destroy_import(export->exp_imp_reverse); } /* for the rest part, we return -ENOTCONN in case of errors @@ -953,13 +1015,17 @@ dont_check_exports: revimp->imp_dlm_fake = 1; revimp->imp_state = LUSTRE_IMP_FULL; - if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_NEXT_VER) { - revimp->imp_msg_magic = LUSTRE_MSG_MAGIC_V2; - lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_NEXT_VER); - } + /* unknown versions will be caught in + * ptlrpc_handle_server_req_in->lustre_unpack_msg() */ + revimp->imp_msg_magic = req->rq_reqmsg->lm_magic; + + if ((export->exp_connect_flags & OBD_CONNECT_AT) && + (revimp->imp_msg_magic != LUSTRE_MSG_MAGIC_V1)) + revimp->imp_msghdr_flags |= MSGHDR_AT_SUPPORT; + else + revimp->imp_msghdr_flags &= ~MSGHDR_AT_SUPPORT; - rc = sptlrpc_import_sec_adapt(revimp, req->rq_svc_ctx, - req->rq_flvr.sf_rpc); + rc = sptlrpc_import_sec_adapt(revimp, req->rq_svc_ctx, &req->rq_flvr); if (rc) { CERROR("Failed to get sec for reverse import: %d\n", rc); export->exp_imp_reverse = NULL; @@ -974,7 +1040,7 @@ out: spin_unlock(&export->exp_lock); } if (targref) - class_decref(targref); + class_decref(targref, __FUNCTION__, cfs_current()); if (rc) req->rq_status = rc; RETURN(rc); @@ -1000,7 +1066,7 @@ void target_destroy_export(struct obd_export *exp) /* exports created from last_rcvd data, and "fake" exports created by lctl don't have an import */ if (exp->exp_imp_reverse != NULL) - destroy_import(exp->exp_imp_reverse); + client_destroy_import(exp->exp_imp_reverse); /* We cancel locks at disconnect time, but this will catch any locks * granted in a race with recovery-induced disconnect. */ @@ -1045,8 +1111,9 @@ struct ptlrpc_request *ptlrpc_clone_req( struct ptlrpc_request *orig_req) copy_req->rq_reqmsg = copy_reqmsg; copy_req->rq_user_desc = udesc; - class_export_get(copy_req->rq_export); + class_export_rpc_get(copy_req->rq_export); CFS_INIT_LIST_HEAD(©_req->rq_list); + CFS_INIT_LIST_HEAD(©_req->rq_replay_list); sptlrpc_svc_ctx_addref(copy_req); if (copy_req->rq_reply_state) { @@ -1060,16 +1127,14 @@ struct ptlrpc_request *ptlrpc_clone_req( struct ptlrpc_request *orig_req) return copy_req; } -void ptlrpc_free_clone( struct ptlrpc_request *req) +void ptlrpc_free_clone(struct ptlrpc_request *req) { - if (req->rq_reply_state) { - ptlrpc_rs_decref(req->rq_reply_state); - req->rq_reply_state = NULL; - } + LASSERT(list_empty(&req->rq_replay_list)); + ptlrpc_req_drop_rs(req); sptlrpc_svc_ctx_decref(req); - class_export_put(req->rq_export); - list_del(&req->rq_list); + class_export_rpc_put(req->rq_export); + list_del_init(&req->rq_list); if (req->rq_user_desc) { int ngroups = req->rq_user_desc->pud_ngroups; @@ -1079,6 +1144,48 @@ void ptlrpc_free_clone( struct ptlrpc_request *req) OBD_FREE_PTR(req); } +static int target_exp_enqueue_req_replay(struct ptlrpc_request *req) +{ + __u64 transno = lustre_msg_get_transno(req->rq_reqmsg); + struct obd_export *exp = req->rq_export; + struct ptlrpc_request *reqiter; + int dup = 0; + + LASSERT(exp); + + spin_lock(&exp->exp_lock); + list_for_each_entry(reqiter, &exp->exp_req_replay_queue, + rq_replay_list) { + if (lustre_msg_get_transno(reqiter->rq_reqmsg) == transno) { + dup = 1; + break; + } + } + + if (dup) { + /* we expect it with RESENT and REPLAY flags */ + if ((lustre_msg_get_flags(req->rq_reqmsg) & + (MSG_RESENT | MSG_REPLAY)) != (MSG_RESENT | MSG_REPLAY)) + CERROR("invalid flags %x of resent replay\n", + lustre_msg_get_flags(req->rq_reqmsg)); + } else { + list_add_tail(&req->rq_replay_list, &exp->exp_req_replay_queue); + } + + spin_unlock(&exp->exp_lock); + return dup; +} + +static void target_exp_dequeue_req_replay(struct ptlrpc_request *req) +{ + LASSERT(!list_empty(&req->rq_replay_list)); + LASSERT(req->rq_export); + + spin_lock(&req->rq_export->exp_lock); + list_del_init(&req->rq_replay_list); + spin_unlock(&req->rq_export->exp_lock); +} + #ifdef __KERNEL__ static void target_finish_recovery(struct obd_device *obd) { @@ -1087,29 +1194,51 @@ static void target_finish_recovery(struct obd_device *obd) obd->obd_name); ldlm_reprocess_all_ns(obd->obd_namespace); + spin_lock_bh(&obd->obd_processing_task_lock); + if (list_empty(&obd->obd_req_replay_queue) && + list_empty(&obd->obd_lock_replay_queue) && + list_empty(&obd->obd_final_req_queue)) { + obd->obd_processing_task = 0; + } else { + CERROR("%s: Recovery queues ( %s%s%s) are not empty\n", + obd->obd_name, + list_empty(&obd->obd_req_replay_queue) ? "" : "req ", + list_empty(&obd->obd_lock_replay_queue) ? "" : "lock ", + list_empty(&obd->obd_final_req_queue) ? "" : "final "); + spin_unlock_bh(&obd->obd_processing_task_lock); + LBUG(); + } + spin_unlock_bh(&obd->obd_processing_task_lock); /* when recovery finished, cleanup orphans on mds and ost */ if (OBT(obd) && OBP(obd, postrecov)) { int rc = OBP(obd, postrecov)(obd); - LCONSOLE_WARN("%s: recovery %s: rc %d\n", obd->obd_name, - rc < 0 ? "failed" : "complete", rc); + if (rc < 0) + LCONSOLE_WARN("%s: Post recovery failed, rc %d\n", + obd->obd_name, rc); } - obd->obd_recovery_end = CURRENT_SECONDS; + obd->obd_recovery_end = cfs_time_current_sec(); EXIT; } static void abort_req_replay_queue(struct obd_device *obd) { struct ptlrpc_request *req, *n; + struct list_head abort_list; - list_for_each_entry_safe(req, n, &obd->obd_req_replay_queue, rq_list) { + CFS_INIT_LIST_HEAD(&abort_list); + spin_lock_bh(&obd->obd_processing_task_lock); + list_splice_init(&obd->obd_req_replay_queue, &abort_list); + spin_unlock_bh(&obd->obd_processing_task_lock); + list_for_each_entry_safe(req, n, &abort_list, rq_list) { DEBUG_REQ(D_WARNING, req, "aborted:"); req->rq_status = -ENOTCONN; if (ptlrpc_error(req)) { DEBUG_REQ(D_ERROR, req, "failed abort_req_reply; skipping"); } + target_exp_dequeue_req_replay(req); ptlrpc_free_clone(req); } } @@ -1117,8 +1246,13 @@ 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_for_each_entry_safe(req, n, &obd->obd_lock_replay_queue, rq_list){ + CFS_INIT_LIST_HEAD(&abort_list); + spin_lock_bh(&obd->obd_processing_task_lock); + list_splice_init(&obd->obd_lock_replay_queue, &abort_list); + spin_unlock_bh(&obd->obd_processing_task_lock); + list_for_each_entry_safe(req, n, &abort_list, rq_list){ DEBUG_REQ(D_ERROR, req, "aborted:"); req->rq_status = -ENOTCONN; if (ptlrpc_error(req)) { @@ -1142,10 +1276,12 @@ 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; ENTRY; LASSERT(obd->obd_stopping); + CFS_INIT_LIST_HEAD(&clean_list); spin_lock_bh(&obd->obd_processing_task_lock); if (!obd->obd_recovering) { spin_unlock_bh(&obd->obd_processing_task_lock); @@ -1154,36 +1290,29 @@ void target_cleanup_recovery(struct obd_device *obd) } obd->obd_recovering = obd->obd_abort_recovery = 0; target_cancel_recovery_timer(obd); + + list_splice_init(&obd->obd_req_replay_queue, &clean_list); spin_unlock_bh(&obd->obd_processing_task_lock); - list_for_each_entry_safe(req, n, &obd->obd_req_replay_queue, rq_list) { - LASSERT (req->rq_reply_state == 0); - ptlrpc_free_clone(req); - } - list_for_each_entry_safe(req, n, &obd->obd_lock_replay_queue, rq_list){ - LASSERT (req->rq_reply_state == 0); + list_for_each_entry_safe(req, n, &clean_list, rq_list) { + LASSERT(req->rq_reply_state == 0); + target_exp_dequeue_req_replay(req); ptlrpc_free_clone(req); } - list_for_each_entry_safe(req, n, &obd->obd_final_req_queue, rq_list) { - LASSERT (req->rq_reply_state == 0); + + spin_lock_bh(&obd->obd_processing_task_lock); + list_splice_init(&obd->obd_lock_replay_queue, &clean_list); + list_splice_init(&obd->obd_final_req_queue, &clean_list); + spin_unlock_bh(&obd->obd_processing_task_lock); + + list_for_each_entry_safe(req, n, &clean_list, rq_list){ + LASSERT(req->rq_reply_state == 0); ptlrpc_free_clone(req); } EXIT; } -static void target_recovery_expired(unsigned long castmeharder) -{ - struct obd_device *obd = (struct obd_device *)castmeharder; - CERROR("%s: recovery timed out, aborting\n", obd->obd_name); - spin_lock_bh(&obd->obd_processing_task_lock); - if (obd->obd_recovering) - obd->obd_abort_recovery = 1; - cfs_waitq_signal(&obd->obd_next_transno_waitq); - spin_unlock_bh(&obd->obd_processing_task_lock); -} - - /* obd_processing_task_lock should be held */ void target_cancel_recovery_timer(struct obd_device *obd) { @@ -1191,42 +1320,101 @@ void target_cancel_recovery_timer(struct obd_device *obd) cfs_timer_disarm(&obd->obd_recovery_timer); } -static void reset_recovery_timer(struct obd_device *obd) +/* extend = 1 means require at least "duration" seconds left in the timer, + extend = 0 means set the total duration (start_recovery_timer) */ +static void reset_recovery_timer(struct obd_device *obd, int duration, + int extend) { - time_t timeout_shift = OBD_RECOVERY_TIMEOUT; + cfs_time_t now = cfs_time_current_sec(); + cfs_duration_t left; + spin_lock_bh(&obd->obd_processing_task_lock); - if (!obd->obd_recovering) { + if (!obd->obd_recovering || obd->obd_abort_recovery) { spin_unlock_bh(&obd->obd_processing_task_lock); return; } - if (cfs_time_current_sec() + OBD_RECOVERY_TIMEOUT > - obd->obd_recovery_start + obd->obd_recovery_max_time) - timeout_shift = obd->obd_recovery_start + - obd->obd_recovery_max_time - cfs_time_current_sec(); - cfs_timer_arm(&obd->obd_recovery_timer, cfs_time_shift(timeout_shift)); + + left = cfs_time_sub(obd->obd_recovery_end, now); + + if (extend && (duration > left)) + obd->obd_recovery_timeout += duration - left; + else if (!extend && (duration > obd->obd_recovery_timeout)) + /* Track the client's largest expected replay time */ + obd->obd_recovery_timeout = duration; +#ifdef CRAY_XT3 + /* + * If total recovery time already exceed the + * obd_recovery_max_time, then CRAY XT3 will + * abort the recovery + */ + if(obd->obd_recovery_timeout > obd->obd_recovery_max_time) + obd->obd_recovery_timeout = obd->obd_recovery_max_time; +#endif + obd->obd_recovery_end = obd->obd_recovery_start + + obd->obd_recovery_timeout; + if (!cfs_timer_is_armed(&obd->obd_recovery_timer) || + cfs_time_before(now, obd->obd_recovery_end)) { + left = cfs_time_sub(obd->obd_recovery_end, now); + cfs_timer_arm(&obd->obd_recovery_timer, cfs_time_shift(left)); + } spin_unlock_bh(&obd->obd_processing_task_lock); - CDEBUG(D_HA, "%s: timer will expire in %u seconds\n", obd->obd_name, - (unsigned int)timeout_shift); - /* Only used for lprocfs_status */ - obd->obd_recovery_end = CURRENT_SECONDS + timeout_shift; + CDEBUG(D_HA, "%s: recovery timer will expire in %u seconds\n", + obd->obd_name, (unsigned)left); } - -/* Only start it the first time called */ -void target_start_recovery_timer(struct obd_device *obd) +static void check_and_start_recovery_timer(struct obd_device *obd) { spin_lock_bh(&obd->obd_processing_task_lock); - if (obd->obd_recovery_handler - || timer_pending((struct timer_list *)&obd->obd_recovery_timer)) { + if (cfs_timer_is_armed(&obd->obd_recovery_timer)) { spin_unlock_bh(&obd->obd_processing_task_lock); return; } - CWARN("%s: starting recovery timer (%us)\n", obd->obd_name, - OBD_RECOVERY_TIMEOUT); - cfs_timer_init(&obd->obd_recovery_timer, target_recovery_expired, obd); + CDEBUG(D_HA, "%s: starting recovery timer\n", obd->obd_name); + obd->obd_recovery_start = cfs_time_current_sec(); + /* minimum */ + obd->obd_recovery_timeout = OBD_RECOVERY_FACTOR * obd_timeout; spin_unlock_bh(&obd->obd_processing_task_lock); - reset_recovery_timer(obd); + reset_recovery_timer(obd, obd->obd_recovery_timeout, 0); +} + +/* Reset the timer with each new client connection */ +/* + * This timer is actually reconnect_timer, which is for making sure + * the total recovery window is at least as big as my reconnect + * attempt timing. So the initial recovery time_out will be set to + * OBD_RECOVERY_FACTOR * obd_timeout. If the timeout coming + * from client is bigger than this, then the recovery time_out will + * be extend to make sure the client could be reconnected, in the + * process, the timeout from the new client should be ignored. + */ + +static void +target_start_and_reset_recovery_timer(struct obd_device *obd, + struct ptlrpc_request *req, + int new_client) +{ + int service_time = lustre_msg_get_service_time(req->rq_reqmsg); + + if (!new_client && service_time) + /* Teach server about old server's estimates, as first guess + * at how long new requests will take. */ + at_add(&req->rq_rqbd->rqbd_service->srv_at_estimate, + service_time); + + check_and_start_recovery_timer(obd); + + /* convert the service time to rpc timeout, + * reuse service_time to limit stack usage */ + service_time = at_est2timeout(service_time); + + /* 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 */ + service_time += 2 * (CONNECTION_SWITCH_MAX + CONNECTION_SWITCH_INC + + INITIAL_CONNECT_TIMEOUT); + if (service_time > obd->obd_recovery_timeout && !new_client) + reset_recovery_timer(obd, service_time, 0); } #ifdef __KERNEL__ @@ -1279,12 +1467,18 @@ static int check_for_next_transno(struct obd_device *obd) next_transno, queue_len, completed, connected, req_transno); obd->obd_next_recovery_transno = req_transno; wake_up = 1; + } else if (OBD_FAIL_CHECK(OBD_FAIL_MDS_RECOVERY_ACCEPTS_GAPS)) { + CDEBUG(D_HA, "accepting transno gaps is explicitly allowed" + " by fail_lock, waking up ("LPD64")\n", next_transno); + obd->obd_next_recovery_transno = req_transno; + wake_up = 1; } else if (queue_len == atomic_read(&obd->obd_req_replay_clients)) { /* some clients haven't connected in time, but we can try * to replay requests that demand on already committed ones * also, we can replay first non-committed transation */ LASSERT(req_transno != 0); - if (req_transno == obd->obd_last_committed + 1) { + if (obd->obd_version_recov || + req_transno == obd->obd_last_committed + 1) { obd->obd_next_recovery_transno = req_transno; } else if (req_transno > obd->obd_last_committed) { /* can't continue recovery: have no needed transno */ @@ -1382,6 +1576,11 @@ static struct ptlrpc_request *target_next_final_ping(struct obd_device *obd) req = list_entry(obd->obd_final_req_queue.next, struct ptlrpc_request, rq_list); list_del_init(&req->rq_list); + if (req->rq_export->exp_in_recovery) { + spin_lock(&req->rq_export->exp_lock); + req->rq_export->exp_in_recovery = 0; + spin_unlock(&req->rq_export->exp_lock); + } } else { req = NULL; } @@ -1389,6 +1588,11 @@ static struct ptlrpc_request *target_next_final_ping(struct obd_device *obd) return req; } +static inline int req_vbr_done(struct obd_export *exp) +{ + return (exp->exp_vbr_failed == 0); +} + static inline int req_replay_done(struct obd_export *exp) { return (exp->exp_req_replay_needed == 0); @@ -1406,7 +1610,7 @@ static inline int connect_done(struct obd_export *exp) static int check_for_clients(struct obd_device *obd) { - if (obd->obd_abort_recovery) + if (obd->obd_abort_recovery || obd->obd_version_recov) return 1; LASSERT(obd->obd_connected_clients <= obd->obd_max_recoverable_clients); if (obd->obd_no_conn == 0 && @@ -1442,14 +1646,28 @@ static int handle_recovery_req(struct ptlrpc_thread *thread, /* don't reset timer for final stage */ if (!req_replay_done(req->rq_export) || !lock_replay_done(req->rq_export)) - reset_recovery_timer(class_exp2obd(req->rq_export)); + reset_recovery_timer(class_exp2obd(req->rq_export), + AT_OFF ? obd_timeout : + at_get(&req->rq_rqbd->rqbd_service->srv_at_estimate), 1); + + /** + * bz18031: increase next_recovery_transno before ptlrpc_free_clone() + * will drop exp_rpc reference + */ + if (!req_replay_done(req->rq_export)) { + spin_lock_bh(&req->rq_export->exp_obd->obd_processing_task_lock); + req->rq_export->exp_obd->obd_next_recovery_transno++; + spin_unlock_bh(&req->rq_export->exp_obd->obd_processing_task_lock); + target_exp_dequeue_req_replay(req); + } ptlrpc_free_clone(req); RETURN(0); } static int target_recovery_thread(void *arg) { - struct obd_device *obd = arg; + struct lu_target *lut = arg; + struct obd_device *obd = lut->lut_obd; struct ptlrpc_request *req; struct target_recovery_data *trd = &obd->obd_recovery_data; struct l_wait_info lwi = { 0 }; @@ -1457,7 +1675,6 @@ static int target_recovery_thread(void *arg) unsigned long flags; struct lu_env env; struct ptlrpc_thread fake_svc_thread, *thread = &fake_svc_thread; - __u32 recov_ctx_tags = LCT_MD_THREAD; int rc = 0; ENTRY; @@ -1468,7 +1685,7 @@ static int target_recovery_thread(void *arg) RECALC_SIGPENDING; SIGNAL_MASK_UNLOCK(current, flags); - rc = lu_context_init(&env.le_ctx, recov_ctx_tags); + rc = lu_context_init(&env.le_ctx, LCT_MD_THREAD); if (rc) RETURN(rc); @@ -1476,8 +1693,8 @@ static int target_recovery_thread(void *arg) env.le_ctx.lc_thread = thread; CERROR("%s: started recovery thread pid %d\n", obd->obd_name, - current->pid); - trd->trd_processing_task = current->pid; + cfs_curproc_pid()); + trd->trd_processing_task = cfs_curproc_pid(); obd->obd_recovering = 1; complete(&trd->trd_starting); @@ -1487,81 +1704,87 @@ static int target_recovery_thread(void *arg) l_wait_event(obd->obd_next_transno_waitq, check_for_clients(obd), &lwi); - spin_lock_bh(&obd->obd_processing_task_lock); - target_cancel_recovery_timer(obd); - spin_unlock_bh(&obd->obd_processing_task_lock); - /* If some clients haven't connected in time, evict them */ - if (obd->obd_abort_recovery) { + if (obd->obd_connected_clients < obd->obd_max_recoverable_clients) { CWARN("Some clients haven't connect in time (%d/%d)," "evict them\n", obd->obd_connected_clients, obd->obd_max_recoverable_clients); - obd->obd_abort_recovery = obd->obd_stopping; - class_disconnect_stale_exports(obd, connect_done); + class_disconnect_stale_exports(obd, connect_done, + exp_flags_from_obd(obd) | + OBD_OPT_ABORT_RECOV); + /** + * if recovery proceeds with versions then some clients may be + * timed out waiting for others and trying to reconnect. + * Extend timer for such reconnect cases. + */ + if (obd->obd_version_recov) + reset_recovery_timer(obd, RECONNECT_DELAY_MAX * 2, 1); } + /* next stage: replay requests */ delta = jiffies; obd->obd_req_replaying = 1; CDEBUG(D_INFO, "1: request replay stage - %d clients from t"LPU64"\n", - atomic_read(&obd->obd_req_replay_clients), - obd->obd_next_recovery_transno); + atomic_read(&obd->obd_req_replay_clients), + obd->obd_next_recovery_transno); while ((req = target_next_replay_req(obd))) { - LASSERT(trd->trd_processing_task == current->pid); + LASSERT(trd->trd_processing_task == cfs_curproc_pid()); DEBUG_REQ(D_HA, req, "processing t"LPD64" from %s", lustre_msg_get_transno(req->rq_reqmsg), libcfs_nid2str(req->rq_peer.nid)); handle_recovery_req(thread, req, trd->trd_recovery_handler); obd->obd_replayed_requests++; - spin_lock_bh(&obd->obd_processing_task_lock); - obd->obd_next_recovery_transno++; - spin_unlock_bh(&obd->obd_processing_task_lock); } - spin_lock_bh(&obd->obd_processing_task_lock); - target_cancel_recovery_timer(obd); - spin_unlock_bh(&obd->obd_processing_task_lock); - /* If some clients haven't replayed requests in time, evict them */ if (obd->obd_abort_recovery) { - CDEBUG(D_ERROR, "req replay timed out, aborting ...\n"); - obd->obd_abort_recovery = obd->obd_stopping; - class_disconnect_stale_exports(obd, req_replay_done); + CDEBUG(D_WARNING, "req replay is aborted\n"); + class_disconnect_stale_exports(obd, req_replay_done, + exp_flags_from_obd(obd) | + OBD_OPT_ABORT_RECOV); abort_req_replay_queue(obd); } + LASSERT(list_empty(&obd->obd_req_replay_queue)); + /* The second stage: replay locks */ CDEBUG(D_INFO, "2: lock replay stage - %d clients\n", atomic_read(&obd->obd_lock_replay_clients)); while ((req = target_next_replay_lock(obd))) { - LASSERT(trd->trd_processing_task == current->pid); - DEBUG_REQ(D_HA|D_WARNING, req, "processing lock from %s: ", + LASSERT(trd->trd_processing_task == cfs_curproc_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); obd->obd_replayed_locks++; } - spin_lock_bh(&obd->obd_processing_task_lock); - target_cancel_recovery_timer(obd); - spin_unlock_bh(&obd->obd_processing_task_lock); /* If some clients haven't replayed requests in time, evict them */ if (obd->obd_abort_recovery) { - int stale; - CERROR("lock replay timed out, aborting ...\n"); - obd->obd_abort_recovery = obd->obd_stopping; - stale = class_disconnect_stale_exports(obd, lock_replay_done); + CERROR("lock replay is aborted\n"); + class_disconnect_stale_exports(obd, lock_replay_done, + exp_flags_from_obd(obd) | + OBD_OPT_ABORT_RECOV); abort_lock_replay_queue(obd); } + LASSERT(list_empty(&obd->obd_lock_replay_queue)); + /* The third stage: reply on final pings */ + CDEBUG(D_INFO, "3: final stage - process recovery completion pings\n"); + /** evict exports failed VBR */ + class_disconnect_stale_exports(obd, req_vbr_done, + exp_flags_from_obd(obd) | + OBD_OPT_ABORT_RECOV); + /** Update server last boot epoch */ + lut_boot_epoch_update(lut); /* We drop recoverying flag to forward all new requests * to regular mds_handle() since now */ spin_lock_bh(&obd->obd_processing_task_lock); obd->obd_recovering = obd->obd_abort_recovery = 0; + target_cancel_recovery_timer(obd); spin_unlock_bh(&obd->obd_processing_task_lock); - /* The third stage: reply on final pings */ - CDEBUG(D_INFO, "3: final stage - process recovery completion pings\n"); while ((req = target_next_final_ping(obd))) { - LASSERT(trd->trd_processing_task == current->pid); + LASSERT(trd->trd_processing_task == cfs_curproc_pid()); DEBUG_REQ(D_HA, req, "processing final ping from %s: ", libcfs_nid2str(req->rq_peer.nid)); handle_recovery_req(thread, req, @@ -1573,7 +1796,7 @@ static int target_recovery_thread(void *arg) delta, obd->obd_replayed_requests, obd->obd_replayed_locks); LASSERT(atomic_read(&obd->obd_req_replay_clients) == 0); LASSERT(atomic_read(&obd->obd_lock_replay_clients) == 0); - if (delta > obd_timeout * 2) { + if (delta > obd_timeout * OBD_RECOVERY_FACTOR) { CWARN("too long recovery - read logs\n"); libcfs_debug_dumplog(); } @@ -1586,8 +1809,10 @@ static int target_recovery_thread(void *arg) RETURN(rc); } -int target_start_recovery_thread(struct obd_device *obd, svc_handler_t handler) +static int target_start_recovery_thread(struct lu_target *lut, + svc_handler_t handler) { + struct obd_device *obd = lut->lut_obd; int rc = 0; struct target_recovery_data *trd = &obd->obd_recovery_data; @@ -1596,7 +1821,7 @@ int target_start_recovery_thread(struct obd_device *obd, svc_handler_t handler) init_completion(&trd->trd_finishing); trd->trd_recovery_handler = handler; - if (kernel_thread(target_recovery_thread, obd, 0) > 0) { + if (kernel_thread(target_recovery_thread, lut, 0) > 0) { wait_for_completion(&trd->trd_starting); LASSERT(obd->obd_recovering != 0); } else @@ -1612,7 +1837,7 @@ void target_stop_recovery_thread(struct obd_device *obd) struct target_recovery_data *trd = &obd->obd_recovery_data; CERROR("%s: Aborting recovery\n", obd->obd_name); obd->obd_abort_recovery = 1; - wake_up(&obd->obd_next_transno_waitq); + cfs_waitq_signal(&obd->obd_next_transno_waitq); spin_unlock_bh(&obd->obd_processing_task_lock); wait_for_completion(&trd->trd_finishing); } else { @@ -1628,27 +1853,52 @@ void target_recovery_fini(struct obd_device *obd) } EXPORT_SYMBOL(target_recovery_fini); -void target_recovery_init(struct obd_device *obd, svc_handler_t handler) +static void target_recovery_expired(unsigned long castmeharder) { - if (obd->obd_max_recoverable_clients == 0) + struct obd_device *obd = (struct obd_device *)castmeharder; + CDEBUG(D_HA, "%s: recovery timed out; %d clients never reconnected " + "after %lds (%d clients did)\n", + obd->obd_name, obd->obd_recoverable_clients, + cfs_time_current_sec()- obd->obd_recovery_start, + obd->obd_connected_clients); + + spin_lock_bh(&obd->obd_processing_task_lock); + obd->obd_version_recov = 1; + CDEBUG(D_INFO, "VBR is used for %d clients from t"LPU64"\n", + atomic_read(&obd->obd_req_replay_clients), + obd->obd_next_recovery_transno); + cfs_waitq_signal(&obd->obd_next_transno_waitq); + spin_unlock_bh(&obd->obd_processing_task_lock); +} + +void target_recovery_init(struct lu_target *lut, svc_handler_t handler) +{ + struct obd_device *obd = lut->lut_obd; + if (obd->obd_max_recoverable_clients == 0) { + /** Update server last boot epoch */ + lut_boot_epoch_update(lut); return; + } CWARN("RECOVERY: service %s, %d recoverable clients, " "last_transno "LPU64"\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; - target_start_recovery_thread(obd, handler); - obd->obd_recovery_start = CURRENT_SECONDS; - /* Only used for lprocfs_status */ - obd->obd_recovery_end = obd->obd_recovery_start + OBD_RECOVERY_TIMEOUT; + obd->obd_recovery_start = 0; + obd->obd_recovery_end = 0; + obd->obd_recovery_timeout = OBD_RECOVERY_FACTOR * obd_timeout; /* bz13079: this should be set to desired value for ost but not for mds */ obd->obd_recovery_max_time = OBD_RECOVERY_MAX_TIME; + cfs_timer_init(&obd->obd_recovery_timer, target_recovery_expired, obd); + target_start_recovery_thread(lut, handler); } EXPORT_SYMBOL(target_recovery_init); #endif -int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req) +static int target_process_req_flags(struct obd_device *obd, + struct ptlrpc_request *req) { struct obd_export *exp = req->rq_export; LASSERT(exp != NULL); @@ -1665,7 +1915,6 @@ int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req) obd->obd_recoverable_clients--; if (atomic_read(&obd->obd_req_replay_clients) == 0) CDEBUG(D_HA, "all clients have replayed reqs\n"); - wake_up(&obd->obd_next_transno_waitq); } spin_unlock_bh(&obd->obd_processing_task_lock); } @@ -1681,7 +1930,6 @@ int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req) atomic_dec(&obd->obd_lock_replay_clients); if (atomic_read(&obd->obd_lock_replay_clients) == 0) CDEBUG(D_HA, "all clients have replayed locks\n"); - wake_up(&obd->obd_next_transno_waitq); } spin_unlock_bh(&obd->obd_processing_task_lock); } @@ -1695,7 +1943,6 @@ int target_queue_recovery_request(struct ptlrpc_request *req, struct list_head *tmp; int inserted = 0; __u64 transno = lustre_msg_get_transno(req->rq_reqmsg); - ENTRY; if (obd->obd_recovery_data.trd_processing_task == cfs_curproc_pid()) { @@ -1713,6 +1960,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req, RETURN(-ENOMEM); DEBUG_REQ(D_HA, req, "queue final req"); spin_lock_bh(&obd->obd_processing_task_lock); + cfs_waitq_signal(&obd->obd_next_transno_waitq); if (obd->obd_recovering) list_add_tail(&req->rq_list, &obd->obd_final_req_queue); else { @@ -1734,6 +1982,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req, RETURN(-ENOMEM); DEBUG_REQ(D_HA, req, "queue lock replay req"); spin_lock_bh(&obd->obd_processing_task_lock); + cfs_waitq_signal(&obd->obd_next_transno_waitq); LASSERT(obd->obd_recovering); /* usually due to recovery abort */ if (!req->rq_export->exp_in_recovery) { @@ -1744,7 +1993,6 @@ int target_queue_recovery_request(struct ptlrpc_request *req, LASSERT(req->rq_export->exp_lock_replay_needed); list_add_tail(&req->rq_list, &obd->obd_lock_replay_queue); spin_unlock_bh(&obd->obd_processing_task_lock); - wake_up(&obd->obd_next_transno_waitq); RETURN(0); } @@ -1780,13 +2028,8 @@ int target_queue_recovery_request(struct ptlrpc_request *req, } spin_unlock_bh(&obd->obd_processing_task_lock); - /* A resent, replayed request that is still on the queue; just drop it. - The queued request will handle this. */ - if ((lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT|MSG_REPLAY)) == - (MSG_RESENT | MSG_REPLAY)) { - DEBUG_REQ(D_ERROR, req, "dropping resent queued req"); + if (OBD_FAIL_CHECK(OBD_FAIL_TGT_REPLAY_DROP)) RETURN(0); - } req = ptlrpc_clone_req(req); if (req == NULL) @@ -1801,6 +2044,13 @@ int target_queue_recovery_request(struct ptlrpc_request *req, } LASSERT(req->rq_export->exp_req_replay_needed); + if (target_exp_enqueue_req_replay(req)) { + spin_unlock_bh(&obd->obd_processing_task_lock); + DEBUG_REQ(D_ERROR, req, "dropping resent queued req"); + ptlrpc_free_clone(req); + RETURN(0); + } + /* XXX O(n^2) */ list_for_each(tmp, &obd->obd_req_replay_queue) { struct ptlrpc_request *reqiter = @@ -1811,16 +2061,25 @@ int target_queue_recovery_request(struct ptlrpc_request *req, inserted = 1; break; } + + if (unlikely(lustre_msg_get_transno(reqiter->rq_reqmsg) == + transno)) { + DEBUG_REQ(D_ERROR, req, "dropping replay: transno " + "has been claimed by another client"); + spin_unlock_bh(&obd->obd_processing_task_lock); + target_exp_dequeue_req_replay(req); + ptlrpc_free_clone(req); + RETURN(0); + } } if (!inserted) list_add_tail(&req->rq_list, &obd->obd_req_replay_queue); obd->obd_requests_queued_for_recovery++; - wake_up(&obd->obd_next_transno_waitq); + cfs_waitq_signal(&obd->obd_next_transno_waitq); spin_unlock_bh(&obd->obd_processing_task_lock); RETURN(0); - } struct obd_device * target_req2obd(struct ptlrpc_request *req) @@ -1834,25 +2093,34 @@ static inline struct ldlm_pool *ldlm_exp2pl(struct obd_export *exp) return &exp->exp_obd->obd_namespace->ns_pool; } +/** + * Packs current SLV and Limit into \a req. + */ int target_pack_pool_reply(struct ptlrpc_request *req) { - struct ldlm_pool *pl; + struct obd_device *obd; ENTRY; - - if (!req->rq_export || !req->rq_export->exp_obd || - !exp_connect_lru_resize(req->rq_export)) { + + /* + * Check that we still have all structures alive as this may + * be some late rpc in shutdown time. + */ + if (unlikely(!req->rq_export || !req->rq_export->exp_obd || + !exp_connect_lru_resize(req->rq_export))) { lustre_msg_set_slv(req->rq_repmsg, 0); lustre_msg_set_limit(req->rq_repmsg, 0); RETURN(0); } - pl = ldlm_exp2pl(req->rq_export); + /* + * OBD is alive here as export is alive, which we checked above. + */ + obd = req->rq_export->exp_obd; - spin_lock(&pl->pl_lock); - LASSERT(ldlm_pool_get_slv(pl) != 0 && ldlm_pool_get_limit(pl) != 0); - lustre_msg_set_slv(req->rq_repmsg, ldlm_pool_get_slv(pl)); - lustre_msg_set_limit(req->rq_repmsg, ldlm_pool_get_limit(pl)); - spin_unlock(&pl->pl_lock); + read_lock(&obd->obd_pool_lock); + lustre_msg_set_slv(req->rq_repmsg, obd->obd_pool_slv); + lustre_msg_set_limit(req->rq_repmsg, obd->obd_pool_limit); + read_unlock(&obd->obd_pool_lock); RETURN(0); } @@ -1867,12 +2135,12 @@ int target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id) if (unlikely(rc)) { DEBUG_REQ(D_ERROR, req, "processing error (%d)", rc); req->rq_status = rc; - return (ptlrpc_error(req)); + return (ptlrpc_send_error(req, 1)); } else { DEBUG_REQ(D_NET, req, "sending reply"); } - return (ptlrpc_send_reply(req, 1)); + return (ptlrpc_send_reply(req, PTLRPC_REPLY_MAYBE_DIFFICULT)); } void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) @@ -1882,15 +2150,19 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) struct obd_device *obd; struct obd_export *exp; struct ptlrpc_service *svc; + ENTRY; - if (req->rq_no_reply) + if (req->rq_no_reply) { + EXIT; return; + } svc = req->rq_rqbd->rqbd_service; rs = req->rq_reply_state; if (rs == NULL || !rs->rs_difficult) { /* no notifiers */ target_send_reply_msg (req, rc, fail_id); + EXIT; return; } @@ -1917,27 +2189,27 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) rs->rs_xid = req->rq_xid; rs->rs_transno = req->rq_transno; rs->rs_export = exp; + rs->rs_opc = lustre_msg_get_opc(rs->rs_msg); - spin_lock(&obd->obd_uncommitted_replies_lock); - - if (rs->rs_transno > obd->obd_last_committed) { + spin_lock(&exp->exp_uncommitted_replies_lock); + CDEBUG(D_NET, "rs transno = "LPU64", last committed = "LPU64"\n", + rs->rs_transno, exp->exp_last_committed); + if (rs->rs_transno > exp->exp_last_committed) { /* not committed already */ - list_add_tail (&rs->rs_obd_list, - &obd->obd_uncommitted_replies); + list_add_tail(&rs->rs_obd_list, + &exp->exp_uncommitted_replies); } + spin_unlock (&exp->exp_uncommitted_replies_lock); - spin_unlock (&obd->obd_uncommitted_replies_lock); - spin_lock (&exp->exp_lock); - - list_add_tail (&rs->rs_exp_list, &exp->exp_outstanding_replies); - + spin_lock(&exp->exp_lock); + list_add_tail(&rs->rs_exp_list, &exp->exp_outstanding_replies); spin_unlock(&exp->exp_lock); netrc = target_send_reply_msg (req, rc, fail_id); spin_lock(&svc->srv_lock); - svc->srv_n_difficult_replies++; + atomic_inc(&svc->srv_n_difficult_replies); if (netrc != 0) { /* error sending: reply is off the net. Also we need +1 @@ -1950,17 +2222,20 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) atomic_inc (&svc->srv_outstanding_replies); } - if (!rs->rs_on_net || /* some notifier */ - list_empty(&rs->rs_exp_list) || /* completed already */ - list_empty(&rs->rs_obd_list)) { - list_add_tail (&rs->rs_list, &svc->srv_reply_queue); - cfs_waitq_signal (&svc->srv_waitq); + spin_lock(&rs->rs_lock); + if (rs->rs_transno <= exp->exp_last_committed || + (!rs->rs_on_net && !rs->rs_no_ack) || + list_empty(&rs->rs_exp_list) || /* completed already */ + list_empty(&rs->rs_obd_list)) { + CDEBUG(D_HA, "Schedule reply immediately\n"); + ptlrpc_dispatch_difficult_reply(rs); } else { list_add (&rs->rs_list, &svc->srv_active_replies); rs->rs_scheduled = 0; /* allow notifier to schedule */ } - + spin_unlock(&rs->rs_lock); spin_unlock(&svc->srv_lock); + EXIT; } int target_handle_ping(struct ptlrpc_request *req) @@ -1971,54 +2246,52 @@ int target_handle_ping(struct ptlrpc_request *req) void target_committed_to_req(struct ptlrpc_request *req) { - struct obd_device *obd; - - if (req == NULL || req->rq_export == NULL) - return; - - obd = req->rq_export->exp_obd; - if (obd == NULL) - return; + struct obd_export *exp = req->rq_export; - if (!obd->obd_no_transno && req->rq_repmsg != NULL) + if (!exp->exp_obd->obd_no_transno && req->rq_repmsg != NULL) lustre_msg_set_last_committed(req->rq_repmsg, - obd->obd_last_committed); + exp->exp_last_committed); else - DEBUG_REQ(D_IOCTL, req, "not sending last_committed update"); + DEBUG_REQ(D_IOCTL, req, "not sending last_committed update (%d/" + "%d)", exp->exp_obd->obd_no_transno, + req->rq_repmsg == NULL); CDEBUG(D_INFO, "last_committed "LPU64", transno "LPU64", xid "LPU64"\n", - obd->obd_last_committed, req->rq_transno, req->rq_xid); + exp->exp_last_committed, req->rq_transno, req->rq_xid); } - EXPORT_SYMBOL(target_committed_to_req); -#ifdef HAVE_QUOTA_SUPPORT int target_handle_qc_callback(struct ptlrpc_request *req) { struct obd_quotactl *oqctl; struct client_obd *cli = &req->rq_export->exp_obd->u.cli; oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL); - if (oqctl == NULL) + if (oqctl == NULL) { + CERROR("Can't unpack obd_quotactl\n"); RETURN(-EPROTO); + } cli->cl_qchk_stat = oqctl->qc_stat; return 0; } +#ifdef HAVE_QUOTA_SUPPORT int target_handle_dqacq_callback(struct ptlrpc_request *req) { #ifdef __KERNEL__ struct obd_device *obd = req->rq_export->exp_obd; struct obd_device *master_obd; + struct obd_device_target *obt; struct lustre_quota_ctxt *qctxt; - struct qunit_data *qdata; - void* rep; - struct qunit_data_old *qdata_old; + struct qunit_data *qdata = NULL; int rc = 0; ENTRY; + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DROP_QUOTA_REQ)) + RETURN(rc); + rc = req_capsule_server_pack(&req->rq_pill); if (rc) { CERROR("packing reply failed!: rc = %d\n", rc); @@ -2027,53 +2300,69 @@ int target_handle_dqacq_callback(struct ptlrpc_request *req) LASSERT(req->rq_export); - /* fixed for bug10707 */ - if ((req->rq_export->exp_connect_flags & OBD_CONNECT_QUOTA64) && - !OBD_FAIL_CHECK(OBD_FAIL_QUOTA_QD_COUNT_32BIT)) { - CDEBUG(D_QUOTA, "qd_count is 64bit!\n"); - rep = req_capsule_server_get(&req->rq_pill, - &RMF_QUNIT_DATA); - LASSERT(rep); - qdata = req_capsule_client_swab_get(&req->rq_pill, - &RMF_QUNIT_DATA, - (void*)lustre_swab_qdata); - } else { - CDEBUG(D_QUOTA, "qd_count is 32bit!\n"); - rep = req_capsule_server_get(&req->rq_pill, &RMF_QUNIT_DATA); - LASSERT(rep); - qdata_old = req_capsule_client_swab_get(&req->rq_pill, - &RMF_QUNIT_DATA, - (void*)lustre_swab_qdata_old); - qdata = lustre_quota_old_to_new(qdata_old); + qdata = quota_get_qdata(req, QUOTA_REQUEST, QUOTA_EXPORT); + if (IS_ERR(qdata)) { + rc = PTR_ERR(qdata); + CDEBUG(D_ERROR, "Can't unpack qunit_data(rc: %d)\n", rc); + req->rq_status = rc; + GOTO(out, rc); } - if (qdata == NULL) - RETURN(-EPROTO); - /* we use the observer */ - LASSERT(obd->obd_observer && obd->obd_observer->obd_observer); + if (!obd->obd_observer || !obd->obd_observer->obd_observer) { + CERROR("Can't find the observer, it is recovering\n"); + req->rq_status = -EAGAIN; + GOTO(out, rc); + } + master_obd = obd->obd_observer->obd_observer; - qctxt = &master_obd->u.obt.obt_qctxt; + obt = &master_obd->u.obt; + qctxt = &obt->obt_qctxt; + + if (!qctxt->lqc_setup || !qctxt->lqc_valid) { + /* quota_type has not been processed yet, return EAGAIN + * until we know whether or not quotas are supposed to + * be enabled */ + CDEBUG(D_QUOTA, "quota_type not processed yet, return " + "-EAGAIN\n"); + req->rq_status = -EAGAIN; + GOTO(out, rc); + } + + down_read(&obt->obt_rwsem); + if (qctxt->lqc_lqs_hash == NULL) { + up_read(&obt->obt_rwsem); + /* quota_type has not been processed yet, return EAGAIN + * until we know whether or not quotas are supposed to + * be enabled */ + CDEBUG(D_QUOTA, "quota_ctxt is not ready yet, return " + "-EAGAIN\n"); + req->rq_status = -EAGAIN; + GOTO(out, rc); + } LASSERT(qctxt->lqc_handler); rc = qctxt->lqc_handler(master_obd, qdata, lustre_msg_get_opc(req->rq_reqmsg)); + up_read(&obt->obt_rwsem); if (rc && rc != -EDQUOT) CDEBUG(rc == -EBUSY ? D_QUOTA : D_ERROR, - "dqacq failed! (rc:%d)\n", rc); + "dqacq/dqrel failed! (rc:%d)\n", rc); + req->rq_status = rc; - /* the qd_count might be changed in lqc_handler */ - if ((req->rq_export->exp_connect_flags & OBD_CONNECT_QUOTA64) && - !OBD_FAIL_CHECK(OBD_FAIL_QUOTA_QD_COUNT_32BIT)) { - memcpy(rep, qdata, sizeof(*qdata)); - } else { - qdata_old = lustre_quota_new_to_old(qdata); - memcpy(rep, qdata_old, sizeof(*qdata_old)); + rc = quota_copy_qdata(req, qdata, QUOTA_REPLY, QUOTA_EXPORT); + if (rc < 0) { + CERROR("Can't pack qunit_data(rc: %d)\n", rc); + GOTO(out, rc); } - req->rq_status = rc; - rc = ptlrpc_reply(req); - RETURN(rc); + /* Block the quota req. b=14840 */ + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_BLOCK_QUOTA_REQ, obd_timeout); + EXIT; + +out: + rc = ptlrpc_reply(req); + return rc; #else return 0; #endif /* !__KERNEL__ */ @@ -2087,5 +2376,85 @@ ldlm_mode_t lck_compat_array[] = { [LCK_CW] LCK_COMPAT_CW, [LCK_CR] LCK_COMPAT_CR, [LCK_NL] LCK_COMPAT_NL, - [LCK_GROUP] LCK_COMPAT_GROUP + [LCK_GROUP] LCK_COMPAT_GROUP, + [LCK_COS] LCK_COMPAT_COS, }; + +/** + * Rather arbitrary mapping from LDLM error codes to errno values. This should + * not escape to the user level. + */ +int ldlm_error2errno(ldlm_error_t error) +{ + int result; + + switch (error) { + case ELDLM_OK: + result = 0; + break; + case ELDLM_LOCK_CHANGED: + result = -ESTALE; + break; + case ELDLM_LOCK_ABORTED: + result = -ENAVAIL; + break; + case ELDLM_LOCK_REPLACED: + result = -ESRCH; + break; + case ELDLM_NO_LOCK_DATA: + result = -ENOENT; + break; + case ELDLM_NAMESPACE_EXISTS: + result = -EEXIST; + break; + case ELDLM_BAD_NAMESPACE: + result = -EBADF; + break; + default: + if (((int)error) < 0) /* cast to signed type */ + result = error; /* as ldlm_error_t can be unsigned */ + else { + CERROR("Invalid DLM result code: %i\n", error); + result = -EPROTO; + } + } + return result; +} +EXPORT_SYMBOL(ldlm_error2errno); + +/** + * Dual to ldlm_error2errno(): maps errno values back to ldlm_error_t. + */ +ldlm_error_t ldlm_errno2error(int err_no) +{ + int error; + + switch (err_no) { + case 0: + error = ELDLM_OK; + break; + case -ESTALE: + error = ELDLM_LOCK_CHANGED; + break; + case -ENAVAIL: + error = ELDLM_LOCK_ABORTED; + break; + case -ESRCH: + error = ELDLM_LOCK_REPLACED; + break; + case -ENOENT: + error = ELDLM_NO_LOCK_DATA; + break; + case -EEXIST: + error = ELDLM_NAMESPACE_EXISTS; + break; + case -EBADF: + error = ELDLM_BAD_NAMESPACE; + break; + default: + error = err_no; + } + return error; +} +EXPORT_SYMBOL(ldlm_errno2error); +