X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lib.c;h=1d4d6b78ff5c85509585fdfa0d9a9d75285324f8;hp=b92c1de9f29102141f881dec8be6af1a5833c787;hb=892078e3b566c04471e7dcf2c28e66f2f3584f93;hpb=f6995cf04407dff15d6ca79ca44cfa97dc6eb014 diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c index b92c1de..1d4d6b7 100644 --- a/lustre/ldlm/ldlm_lib.c +++ b/lustre/ldlm/ldlm_lib.c @@ -260,17 +260,18 @@ static int osc_on_mdt(char *obdname) */ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) { - struct client_obd *cli = &obddev->u.cli; - struct obd_import *imp; - struct obd_uuid server_uuid; - int rq_portal, rp_portal, connect_op; - char *name = obddev->obd_type->typ_name; - ldlm_ns_type_t ns_type = LDLM_NS_TYPE_UNKNOWN; - int rc; - ENTRY; + struct client_obd *cli = &obddev->u.cli; + struct obd_import *imp; + struct obd_uuid server_uuid; + int rq_portal, rp_portal, connect_op; + 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); + int rc; + ENTRY; - /* In a more perfect world, we would hang a ptlrpc_client off of - * obd_type and just use the values from there. */ + /* In a more perfect world, we would hang a ptlrpc_client off of + * obd_type and just use the values from there. */ if (!strcmp(name, LUSTRE_OSC_NAME)) { rq_portal = OST_REQUEST_PORTAL; rp_portal = OSC_REPLY_PORTAL; @@ -283,7 +284,12 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) rq_portal = MDS_REQUEST_PORTAL; rp_portal = MDC_REPLY_PORTAL; connect_op = MDS_CONNECT; - cli->cl_sp_me = LUSTRE_SP_CLI; + if (is_lwp_on_ost(cli_name)) + cli->cl_sp_me = LUSTRE_SP_OST; + else if (is_lwp_on_mdt(cli_name)) + cli->cl_sp_me = LUSTRE_SP_MDT; + else + cli->cl_sp_me = LUSTRE_SP_CLI; cli->cl_sp_to = LUSTRE_SP_MDT; ns_type = LDLM_NS_TYPE_MDC; } else if (!strcmp(name, LUSTRE_OSP_NAME)) { @@ -301,7 +307,7 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) rq_portal = OST_REQUEST_PORTAL; } rp_portal = OSC_REPLY_PORTAL; - cli->cl_sp_me = LUSTRE_SP_CLI; + cli->cl_sp_me = LUSTRE_SP_MDT; } else if (!strcmp(name, LUSTRE_MGC_NAME)) { rq_portal = MGS_REQUEST_PORTAL; rp_portal = MGC_REPLY_PORTAL; @@ -415,6 +421,23 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) 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); + cli->cl_max_mod_rpcs_in_flight = 0; + cli->cl_mod_rpcs_in_flight = 0; + cli->cl_close_rpcs_in_flight = 0; + init_waitqueue_head(&cli->cl_mod_rpcs_waitq); + cli->cl_mod_tag_bitmap = NULL; + + 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, + BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long)); + if (cli->cl_mod_tag_bitmap == NULL) + GOTO(err, rc = -ENOMEM); + } + rc = ldlm_get_ref(); if (rc) { CERROR("ldlm_get_ref failed: %d\n", rc); @@ -439,10 +462,9 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) GOTO(err_import, rc); } - 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_v3); - cli->cl_max_mds_cookiesize = sizeof(struct llog_cookie); + cli->cl_import = imp; + /* cli->cl_max_mds_easize updated by mdc_init_ea_size() */ + cli->cl_max_mds_easize = sizeof(struct lov_mds_md_v3); if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) { if (!strcmp(lustre_cfg_string(lcfg, 3), "inactive")) { @@ -465,8 +487,6 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) GOTO(err_import, rc = -ENOMEM); } - cli->cl_qchk_stat = CL_NOT_QUOTACHECKED; - RETURN(rc); err_import: @@ -474,6 +494,10 @@ err_import: err_ldlm: ldlm_put_ref(); err: + if (cli->cl_mod_tag_bitmap != NULL) + OBD_FREE(cli->cl_mod_tag_bitmap, + BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long)); + cli->cl_mod_tag_bitmap = NULL; RETURN(rc); } @@ -481,6 +505,7 @@ EXPORT_SYMBOL(client_obd_setup); int client_obd_cleanup(struct obd_device *obddev) { + struct client_obd *cli = &obddev->u.cli; ENTRY; ldlm_namespace_free_post(obddev->obd_namespace); @@ -490,6 +515,12 @@ int client_obd_cleanup(struct obd_device *obddev) LASSERT(obddev->u.cli.cl_import == NULL); ldlm_put_ref(); + + if (cli->cl_mod_tag_bitmap != NULL) + OBD_FREE(cli->cl_mod_tag_bitmap, + BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long)); + cli->cl_mod_tag_bitmap = NULL; + RETURN(0); } EXPORT_SYMBOL(client_obd_cleanup); @@ -505,6 +536,7 @@ 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; @@ -529,6 +561,10 @@ 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; } @@ -544,6 +580,10 @@ int client_connect_import(const struct lu_env *env, ocd->ocd_connect_flags, "old "LPX64", new "LPX64"\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; } ptlrpc_pinger_add_import(imp); @@ -648,8 +688,7 @@ int server_disconnect_export(struct obd_export *exp) if (exp->exp_imp_reverse) ptlrpc_cleanup_imp(exp->exp_imp_reverse); - if (exp->exp_obd->obd_namespace != NULL) - ldlm_cancel_locks_for_export(exp); + ldlm_bl_thread_wakeup(); /* complete all outstanding replies */ spin_lock(&exp->exp_lock); @@ -682,13 +721,13 @@ static int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp, struct obd_uuid *cluuid) { - ENTRY; + struct lustre_handle *hdl; + ENTRY; - if (exp->exp_connection && exp->exp_imp_reverse) { - struct lustre_handle *hdl; + hdl = &exp->exp_imp_reverse->imp_remote_handle; + if (exp->exp_connection && lustre_handle_is_used(hdl)) { struct obd_device *target; - hdl = &exp->exp_imp_reverse->imp_remote_handle; target = exp->exp_obd; /* Might be a re-connect after a partition. */ @@ -758,22 +797,131 @@ static void check_and_start_recovery_timer(struct obd_device *obd, struct ptlrpc_request *req, int new_client); +/** + * update flags for import during reconnect process + */ +static int rev_import_flags_update(struct obd_import *revimp, + struct ptlrpc_request *req) +{ + int rc; + struct obd_connect_data *data; + + data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA); + + if (data->ocd_connect_flags & OBD_CONNECT_AT) + revimp->imp_msghdr_flags |= MSGHDR_AT_SUPPORT; + else + revimp->imp_msghdr_flags &= ~MSGHDR_AT_SUPPORT; + + revimp->imp_msghdr_flags |= MSGHDR_CKSUM_INCOMPAT18; + + rc = sptlrpc_import_sec_adapt(revimp, req->rq_svc_ctx, &req->rq_flvr); + if (rc) { + CERROR("%s: cannot get reverse import %s security: rc = %d\n", + revimp->imp_client->cli_name, + libcfs_id2str(req->rq_peer), rc); + return rc; + } + + return 0; +} + +/** + * Allocate a new reverse import for an export. + * + * \retval -errno in case error hit + * \retval 0 if reverse import correctly init + **/ +int rev_import_init(struct obd_export *export) +{ + struct obd_device *obd = export->exp_obd; + struct obd_import *revimp; + + LASSERT(export->exp_imp_reverse == NULL); + + revimp = class_new_import(obd); + if (revimp == NULL) + return -ENOMEM; + + revimp->imp_remote_handle.cookie = 0ULL; + revimp->imp_client = &obd->obd_ldlm_client; + revimp->imp_dlm_fake = 1; + + /* it is safe to connect import in new state as no sends possible */ + spin_lock(&export->exp_lock); + export->exp_imp_reverse = revimp; + spin_unlock(&export->exp_lock); + class_import_put(revimp); + + return 0; +} +EXPORT_SYMBOL(rev_import_init); + +/** + * Handle reconnect for an export. + * + * \param exp export to handle reconnect process + * \param req client reconnect request + * + * \retval -rc in case securitfy flavor can't be changed + * \retval 0 in case none problems + */ +static int rev_import_reconnect(struct obd_export *exp, + struct ptlrpc_request *req) +{ + struct obd_import *revimp = exp->exp_imp_reverse; + struct lustre_handle *lh; + int rc; + + /* avoid sending a request until import flags are changed */ + ptlrpc_import_enter_resend(revimp); + + if (revimp->imp_connection != NULL) + ptlrpc_connection_put(revimp->imp_connection); + + /* + * client from recovery don't have a handle so we need to take from + * request. it may produce situation when wrong client connected + * to recovery as we trust a client uuid + */ + lh = req_capsule_client_get(&req->rq_pill, &RMF_CONN); + revimp->imp_remote_handle = *lh; + + /* unknown versions will be caught in + * ptlrpc_handle_server_req_in->lustre_unpack_msg() */ + revimp->imp_msg_magic = req->rq_reqmsg->lm_magic; + + revimp->imp_connection = ptlrpc_connection_addref(exp->exp_connection); + + rc = rev_import_flags_update(revimp, req); + if (rc != 0) { + /* it is safe to still be in RECOVERY phase as we are not able + * to setup correct security flavor so requests are not able to + * be delivered correctly */ + return rc; + } + + /* resend all rpc's via new connection */ + return ptlrpc_import_recovery_state_machine(revimp); +} + int target_handle_connect(struct ptlrpc_request *req) { - struct obd_device *target = NULL, *targref = NULL; - struct obd_export *export = NULL; - struct obd_import *revimp; - struct obd_import *tmp_imp = NULL; - struct lustre_handle conn; - struct lustre_handle *tmp; + struct obd_device *target = NULL; + struct obd_export *export = NULL; + /* connect handle - filled from target_handle_reconnect in + * reconnect case */ + struct lustre_handle conn; + struct lustre_handle *tmp; struct obd_uuid tgtuuid; struct obd_uuid cluuid; - struct obd_uuid remote_uuid; char *str; int rc = 0; char *target_start; int target_len; bool mds_conn = false, lw_client = false; + bool mds_mds_conn = false; + bool new_mds_mds_conn = false; struct obd_connect_data *data, *tmpdata; int size, tmpsize; lnet_nid_t *client_nid = NULL; @@ -803,6 +951,11 @@ int target_handle_connect(struct ptlrpc_request *req) } spin_lock(&target->obd_dev_lock); + /* 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); + if (target->obd_stopping || !target->obd_set_up) { spin_unlock(&target->obd_dev_lock); @@ -824,11 +977,6 @@ 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. */ - targref = class_incref(target, __FUNCTION__, current); - target->obd_conn_inprogress++; spin_unlock(&target->obd_dev_lock); @@ -840,21 +988,6 @@ int target_handle_connect(struct ptlrpc_request *req) obd_str2uuid(&cluuid, str); - /* XXX Extract a nettype and format accordingly. */ - switch (sizeof(lnet_nid_t)) { - /* NB the casts only avoid compiler warnings. */ - case 8: - snprintf(remote_uuid.uuid, sizeof remote_uuid, - "NET_"LPX64"_UUID", (__u64)req->rq_peer.nid); - break; - case 4: - snprintf(remote_uuid.uuid, sizeof remote_uuid, - "NET_%x_UUID", (__u32)req->rq_peer.nid); - break; - default: - LBUG(); - } - tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN); if (tmp == NULL) GOTO(out, rc = -EPROTO); @@ -871,6 +1004,20 @@ int target_handle_connect(struct ptlrpc_request *req) if (rc) GOTO(out, rc); +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0) + /* Don't allow clients to connect that are using old 1.8 format + * protocol conventions (LUSTRE_MSG_MAGIC_v1, !MSGHDR_CKSUM_INCOMPAT18, + * ldlm_flock_policy_wire format, MDT_ATTR_xTIME_SET, etc). The + * FULL20 flag should be set on all connections since 2.0, but no + * longer affects behaviour. + * + * Later this check will be disabled and the flag can be retired + * completely once interop with 3.0 is no longer needed. + */ + if (!(data->ocd_connect_flags & OBD_CONNECT_FULL20)) + GOTO(out, rc = -EPROTO); +#endif + if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) { if (data->ocd_version < LUSTRE_VERSION_CODE - LUSTRE_VERSION_ALLOWED_OFFSET || @@ -896,10 +1043,15 @@ int target_handle_connect(struct ptlrpc_request *req) } } + /* Note: lw_client is needed in MDS-MDS failover during update log + * processing, so we needs to allow lw_client to be connected at + * anytime, instead of only the initial connection */ + lw_client = (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) != 0; + if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL) { mds_conn = (data->ocd_connect_flags & OBD_CONNECT_MDS) != 0; - lw_client = (data->ocd_connect_flags & - OBD_CONNECT_LIGHTWEIGHT) != 0; + mds_mds_conn = (data->ocd_connect_flags & + OBD_CONNECT_MDS_MDS) != 0; /* OBD_CONNECT_MNE_SWAB is defined as OBD_CONNECT_MDS_MDS * for Imperative Recovery connection from MGC to MGS. @@ -1051,7 +1203,9 @@ no_export: if (export == NULL) { /* allow lightweight connections during recovery */ - if (target->obd_recovering && !lw_client) { + /* 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 */ @@ -1065,30 +1219,35 @@ no_export: t = cfs_timer_deadline(&target->obd_recovery_timer); t = cfs_time_sub(t, cfs_time_current()); t = cfs_duration_sec(t); - LCONSOLE_WARN("%s: Denying connection for new client " - "%s (at %s), waiting for all %d known " - "clients (%d recovered, %d in progress, " - "and %d evicted) to recover in %d:%.02d\n", + 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", target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), k, c - i, i, s, (int)t / 60, (int)t % 60); - rc = -EBUSY; - } else { + rc = -EBUSY; + } else { dont_check_exports: - rc = obd_connect(req->rq_svc_thread->t_env, - &export, target, &cluuid, data, - client_nid); + rc = obd_connect(req->rq_svc_thread->t_env, + &export, target, &cluuid, data, + client_nid); if (mds_conn && OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG)) lustre_msg_add_op_flags(req->rq_repmsg, - MSG_CONNECT_RECOVERING); - if (rc == 0) - conn.cookie = export->exp_handle.h_cookie; - } - } else { - rc = obd_reconnect(req->rq_svc_thread->t_env, - export, target, &cluuid, data, client_nid); - } + MSG_CONNECT_RECOVERING); + if (rc == 0) { + conn.cookie = export->exp_handle.h_cookie; + rc = rev_import_init(export); + } + + if (mds_mds_conn) + new_mds_mds_conn = true; + } + } else { + rc = obd_reconnect(req->rq_svc_thread->t_env, + export, target, &cluuid, data, client_nid); + } if (rc) GOTO(out, rc); @@ -1108,11 +1267,6 @@ dont_check_exports: memcpy(tmpdata, data, min(tmpsize, size)); } - /* If all else goes well, this is our RPC return code. */ - req->rq_status = 0; - - lustre_msg_set_handle(req->rq_repmsg, &conn); - /* 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. @@ -1159,14 +1313,19 @@ dont_check_exports: ptlrpc_connection_put(export->exp_connection); } - export->exp_connection = ptlrpc_connection_get(req->rq_peer, - req->rq_self, - &remote_uuid); - if (hlist_unhashed(&export->exp_nid_hash)) { - cfs_hash_add(export->exp_obd->obd_nid_hash, - &export->exp_connection->c_peer.nid, - &export->exp_nid_hash); - } + export->exp_connection = ptlrpc_connection_get(req->rq_peer, + req->rq_self, + &cluuid); + if (hlist_unhashed(&export->exp_nid_hash)) + cfs_hash_add(export->exp_obd->obd_nid_hash, + &export->exp_connection->c_peer.nid, + &export->exp_nid_hash); + + lustre_msg_set_handle(req->rq_repmsg, &conn); + + rc = rev_import_reconnect(export, req); + if (rc != 0) + GOTO(out, rc); if (target->obd_recovering && !export->exp_in_recovery && !lw_client) { int has_transno; @@ -1201,6 +1360,14 @@ dont_check_exports: atomic_inc(&target->obd_req_replay_clients); atomic_inc(&target->obd_lock_replay_clients); + /* Note: MDS-MDS connection is allowed to be connected during + * recovery, no matter if the exports needs to be recoveried. + * Because we need retrieve updates logs from all other MDTs. + * So if the MDS-MDS export is new, obd_max_recoverable_clients + * also needs to be increased to match other recovery checking + * condition. */ + if (new_mds_mds_conn) + target->obd_max_recoverable_clients++; if (atomic_inc_return(&target->obd_connected_clients) == target->obd_max_recoverable_clients) wake_up(&target->obd_next_transno_waitq); @@ -1210,59 +1377,7 @@ dont_check_exports: if (target->obd_recovering && !lw_client) lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING); - tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN); - conn = *tmp; - - /* Return -ENOTCONN in case of errors to let client reconnect. */ - revimp = class_new_import(target); - if (revimp == NULL) { - CERROR("fail to alloc new reverse import.\n"); - GOTO(out, rc = -ENOTCONN); - } - - spin_lock(&export->exp_lock); - if (export->exp_imp_reverse != NULL) - /* destroyed import can be still referenced in ctxt */ - tmp_imp = export->exp_imp_reverse; - export->exp_imp_reverse = revimp; - spin_unlock(&export->exp_lock); - - revimp->imp_connection = ptlrpc_connection_addref(export->exp_connection); - revimp->imp_client = &export->exp_obd->obd_ldlm_client; - revimp->imp_remote_handle = conn; - revimp->imp_dlm_fake = 1; - revimp->imp_state = LUSTRE_IMP_FULL; - - /* Unknown versions will be caught in - * ptlrpc_handle_server_req_in->lustre_unpack_msg(). */ - revimp->imp_msg_magic = req->rq_reqmsg->lm_magic; - - if ((data->ocd_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; - - if ((data->ocd_connect_flags & OBD_CONNECT_FULL20) && - (revimp->imp_msg_magic != LUSTRE_MSG_MAGIC_V1)) - revimp->imp_msghdr_flags |= MSGHDR_CKSUM_INCOMPAT18; - else - revimp->imp_msghdr_flags &= ~MSGHDR_CKSUM_INCOMPAT18; - - 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); - spin_lock(&export->exp_lock); - export->exp_imp_reverse = NULL; - spin_unlock(&export->exp_lock); - class_destroy_import(revimp); - } - - class_import_put(revimp); - out: - if (tmp_imp != NULL) - client_destroy_import(tmp_imp); if (export) { spin_lock(&export->exp_lock); export->exp_connecting = 0; @@ -1270,15 +1385,14 @@ out: class_export_put(export); } - if (targref) { + if (target != NULL) { spin_lock(&target->obd_dev_lock); target->obd_conn_inprogress--; spin_unlock(&target->obd_dev_lock); - class_decref(targref, __func__, current); + class_decref(target, __func__, current); } - if (rc) - req->rq_status = rc; + req->rq_status = rc; RETURN(rc); } @@ -1348,6 +1462,7 @@ 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; + struct ptlrpc_request *dup_req = NULL; int dup = 0; LASSERT(exp); @@ -1356,6 +1471,7 @@ static int target_exp_enqueue_req_replay(struct ptlrpc_request *req) list_for_each_entry(reqiter, &exp->exp_req_replay_queue, rq_replay_list) { if (lustre_msg_get_transno(reqiter->rq_reqmsg) == transno) { + dup_req = reqiter; dup = 1; break; } @@ -1367,6 +1483,16 @@ static int target_exp_enqueue_req_replay(struct ptlrpc_request *req) (MSG_RESENT | MSG_REPLAY)) != (MSG_RESENT | MSG_REPLAY)) CERROR("invalid flags %x of resent replay\n", lustre_msg_get_flags(req->rq_reqmsg)); + + if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) { + __u32 new_conn; + + new_conn = lustre_msg_get_conn_cnt(req->rq_reqmsg); + if (new_conn > + lustre_msg_get_conn_cnt(dup_req->rq_reqmsg)) + lustre_msg_set_conn_cnt(dup_req->rq_reqmsg, + new_conn); + } } else { list_add_tail(&req->rq_replay_list, &exp->exp_req_replay_queue); @@ -1386,8 +1512,9 @@ static void target_exp_dequeue_req_replay(struct ptlrpc_request *req) spin_unlock(&req->rq_export->exp_lock); } -static void target_finish_recovery(struct obd_device *obd) +static void target_finish_recovery(struct lu_target *lut) { + struct obd_device *obd = lut->lut_obd; ENTRY; /* Only log a recovery message when recovery has occurred. */ @@ -1420,6 +1547,13 @@ static void target_finish_recovery(struct obd_device *obd) } spin_unlock(&obd->obd_recovery_task_lock); + if (lut->lut_tdtd != NULL && + (!list_empty(&lut->lut_tdtd->tdtd_replay_list) || + !list_empty(&lut->lut_tdtd->tdtd_replay_finish_list))) { + dtrq_list_dump(lut->lut_tdtd, D_ERROR); + dtrq_list_destroy(lut->lut_tdtd); + } + obd->obd_recovery_end = cfs_time_current_sec(); /* When recovery finished, cleanup orphans on MDS and OST. */ @@ -1486,7 +1620,6 @@ void target_cleanup_recovery(struct obd_device *obd) { struct ptlrpc_request *req, *n; struct list_head clean_list; - ENTRY; INIT_LIST_HEAD(&clean_list); spin_lock(&obd->obd_dev_lock); @@ -1496,6 +1629,7 @@ void target_cleanup_recovery(struct obd_device *obd) return; } obd->obd_recovering = obd->obd_abort_recovery = 0; + obd->obd_force_abort_recovery = 0; spin_unlock(&obd->obd_dev_lock); spin_lock(&obd->obd_recovery_task_lock); @@ -1536,7 +1670,8 @@ static void target_start_recovery_timer(struct obd_device *obd) return; spin_lock(&obd->obd_dev_lock); - if (!obd->obd_recovering || obd->obd_abort_recovery) { + if (!obd->obd_recovering || obd->obd_abort_recovery || + obd->obd_force_abort_recovery) { spin_unlock(&obd->obd_dev_lock); return; } @@ -1577,7 +1712,8 @@ static void extend_recovery_timer(struct obd_device *obd, int drt, bool extend) int to; spin_lock(&obd->obd_dev_lock); - if (!obd->obd_recovering || obd->obd_abort_recovery) { + if (!obd->obd_recovering || obd->obd_abort_recovery || + obd->obd_force_abort_recovery) { spin_unlock(&obd->obd_dev_lock); return; } @@ -1679,23 +1815,15 @@ static inline int exp_finished(struct obd_export *exp) return (exp->exp_in_recovery && !exp->exp_lock_replay_needed); } -/** Checking routines for recovery */ -static int check_for_clients(struct obd_device *obd) -{ - unsigned int clnts = atomic_read(&obd->obd_connected_clients); - - if (obd->obd_abort_recovery || obd->obd_recovery_expired) - return 1; - LASSERT(clnts <= obd->obd_max_recoverable_clients); - return (clnts + obd->obd_stale_clients == - obd->obd_max_recoverable_clients); -} - -static int check_for_next_transno(struct obd_device *obd) +static int check_for_next_transno(struct lu_target *lut) { struct ptlrpc_request *req = NULL; + struct obd_device *obd = lut->lut_obd; + struct target_distribute_txn_data *tdtd = lut->lut_tdtd; int wake_up = 0, connected, completed, queue_len; - __u64 next_transno, req_transno; + __u64 req_transno = 0; + __u64 update_transno = 0; + __u64 next_transno = 0; ENTRY; spin_lock(&obd->obd_recovery_task_lock); @@ -1703,10 +1831,11 @@ static int check_for_next_transno(struct obd_device *obd) req = list_entry(obd->obd_req_replay_queue.next, struct ptlrpc_request, rq_list); req_transno = lustre_msg_get_transno(req->rq_reqmsg); - } else { - req_transno = 0; } + if (tdtd != NULL) + update_transno = distribute_txn_get_next_transno(tdtd); + connected = atomic_read(&obd->obd_connected_clients); completed = connected - atomic_read(&obd->obd_req_replay_clients); queue_len = obd->obd_requests_queued_for_recovery; @@ -1717,13 +1846,21 @@ static int check_for_next_transno(struct obd_device *obd) obd->obd_max_recoverable_clients, connected, completed, queue_len, req_transno, next_transno); - if (obd->obd_abort_recovery) { + if (obd->obd_abort_recovery || obd->obd_force_abort_recovery) { CDEBUG(D_HA, "waking for aborted recovery\n"); wake_up = 1; } else if (obd->obd_recovery_expired) { CDEBUG(D_HA, "waking for expired recovery\n"); wake_up = 1; - } else if (req_transno == next_transno) { + } else if (tdtd != NULL && req != NULL && + is_req_replayed_by_update(req)) { + LASSERTF(req_transno < next_transno, "req_transno "LPU64 + "next_transno"LPU64"\n", req_transno, next_transno); + CDEBUG(D_HA, "waking for duplicate req ("LPU64")\n", + req_transno); + wake_up = 1; + } else if (req_transno == next_transno || + (update_transno != 0 && update_transno <= next_transno)) { CDEBUG(D_HA, "waking for next ("LPD64")\n", next_transno); wake_up = 1; } else if (queue_len > 0 && @@ -1739,10 +1876,10 @@ static int check_for_next_transno(struct obd_device *obd) CDEBUG(d_lvl, "%s: waking for gap in transno, VBR is %s (skip: " LPD64", ql: %d, comp: %d, conn: %d, next: "LPD64 - ", last_committed: "LPD64")\n", + ", next_update "LPD64" last_committed: "LPD64")\n", obd->obd_name, obd->obd_version_recov ? "ON" : "OFF", next_transno, queue_len, completed, connected, - req_transno, obd->obd_last_committed); + req_transno, update_transno, obd->obd_last_committed); obd->obd_next_recovery_transno = req_transno; wake_up = 1; } else if (atomic_read(&obd->obd_req_replay_clients) == 0) { @@ -1758,8 +1895,9 @@ static int check_for_next_transno(struct obd_device *obd) return wake_up; } -static int check_for_next_lock(struct obd_device *obd) +static int check_for_next_lock(struct lu_target *lut) { + struct obd_device *obd = lut->lut_obd; int wake_up = 0; spin_lock(&obd->obd_recovery_task_lock); @@ -1769,7 +1907,7 @@ static int check_for_next_lock(struct obd_device *obd) } else if (atomic_read(&obd->obd_lock_replay_clients) == 0) { CDEBUG(D_HA, "waking for completed lock replay\n"); wake_up = 1; - } else if (obd->obd_abort_recovery) { + } else if (obd->obd_abort_recovery || obd->obd_force_abort_recovery) { CDEBUG(D_HA, "waking for aborted recovery\n"); wake_up = 1; } else if (obd->obd_recovery_expired) { @@ -1786,10 +1924,11 @@ static int check_for_next_lock(struct obd_device *obd) * check its status with help of check_routine * evict dead clients via health_check */ -static int target_recovery_overseer(struct obd_device *obd, - int (*check_routine)(struct obd_device *), +static int target_recovery_overseer(struct lu_target *lut, + int (*check_routine)(struct lu_target *), int (*health_check)(struct obd_export *)) { + struct obd_device *obd = lut->lut_obd; repeat: if ((obd->obd_recovery_start != 0) && (cfs_time_current_sec() >= (obd->obd_recovery_start + obd->obd_recovery_time_hard))) { @@ -1798,11 +1937,11 @@ repeat: } while (wait_event_timeout(obd->obd_next_transno_waitq, - check_routine(obd), + check_routine(lut), msecs_to_jiffies(60 * MSEC_PER_SEC)) == 0) /* wait indefinitely for event, but don't trigger watchdog */; - if (obd->obd_abort_recovery) { + if (obd->obd_abort_recovery || obd->obd_force_abort_recovery) { CWARN("recovery is aborted, evict exports in recovery\n"); /** evict exports which didn't finish recovery yet */ class_disconnect_stale_exports(obd, exp_finished); @@ -1829,50 +1968,13 @@ repeat: return 0; } -static struct ptlrpc_request *target_next_replay_req(struct obd_device *obd) -{ - struct ptlrpc_request *req = NULL; - ENTRY; - - CDEBUG(D_HA, "Waiting for transno "LPD64"\n", - obd->obd_next_recovery_transno); - - CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLAY_DELAY2, cfs_fail_val); - /** It is needed to extend recovery window above recovery_time_soft. - * Extending is possible only in the end of recovery window - * (see more details in handle_recovery_req). - */ - CFS_FAIL_TIMEOUT_MS(OBD_FAIL_TGT_REPLAY_DELAY, 300); - - if (target_recovery_overseer(obd, check_for_next_transno, - exp_req_replay_healthy)) { - abort_req_replay_queue(obd); - abort_lock_replay_queue(obd); - } - - spin_lock(&obd->obd_recovery_task_lock); - if (!list_empty(&obd->obd_req_replay_queue)) { - req = list_entry(obd->obd_req_replay_queue.next, - struct ptlrpc_request, rq_list); - list_del_init(&req->rq_list); - obd->obd_requests_queued_for_recovery--; - spin_unlock(&obd->obd_recovery_task_lock); - } else { - spin_unlock(&obd->obd_recovery_task_lock); - LASSERT(list_empty(&obd->obd_req_replay_queue)); - LASSERT(atomic_read(&obd->obd_req_replay_clients) == 0); - /** evict exports failed VBR */ - class_disconnect_stale_exports(obd, exp_vbr_healthy); - } - RETURN(req); -} - -static struct ptlrpc_request *target_next_replay_lock(struct obd_device *obd) +static struct ptlrpc_request *target_next_replay_lock(struct lu_target *lut) { + struct obd_device *obd = lut->lut_obd; struct ptlrpc_request *req = NULL; CDEBUG(D_HA, "Waiting for lock\n"); - if (target_recovery_overseer(obd, check_for_next_lock, + if (target_recovery_overseer(lut, check_for_next_lock, exp_lock_replay_healthy)) abort_lock_replay_queue(obd); @@ -1966,6 +2068,249 @@ static void handle_recovery_req(struct ptlrpc_thread *thread, EXIT; } +/** Checking routines for recovery */ +static int check_for_recovery_ready(struct lu_target *lut) +{ + struct obd_device *obd = lut->lut_obd; + unsigned int clnts = atomic_read(&obd->obd_connected_clients); + + CDEBUG(D_HA, "connected %d stale %d max_recoverable_clients %d" + " abort %d expired %d\n", clnts, obd->obd_stale_clients, + obd->obd_max_recoverable_clients, obd->obd_abort_recovery, + obd->obd_recovery_expired); + + if (obd->obd_force_abort_recovery) + return 1; + + if (!obd->obd_abort_recovery && !obd->obd_recovery_expired) { + LASSERT(clnts <= obd->obd_max_recoverable_clients); + if (clnts + obd->obd_stale_clients < + obd->obd_max_recoverable_clients) + return 0; + } + + if (lut->lut_tdtd != NULL) { + if (!lut->lut_tdtd->tdtd_replay_ready) { + /* Let's extend recovery timer, in case the recovery + * timer expired, and some clients got evicted */ + extend_recovery_timer(obd, obd->obd_recovery_timeout, + true); + return 0; + } else { + dtrq_list_dump(lut->lut_tdtd, D_HA); + } + } + + return 1; +} + +enum { + REQUEST_RECOVERY = 1, + UPDATE_RECOVERY = 2, +}; + +static __u64 get_next_replay_req_transno(struct obd_device *obd) +{ + __u64 transno = 0; + + if (!list_empty(&obd->obd_req_replay_queue)) { + struct ptlrpc_request *req; + + req = list_entry(obd->obd_req_replay_queue.next, + struct ptlrpc_request, rq_list); + transno = lustre_msg_get_transno(req->rq_reqmsg); + } + + return transno; +} +__u64 get_next_transno(struct lu_target *lut, int *type) +{ + struct obd_device *obd = lut->lut_obd; + struct target_distribute_txn_data *tdtd = lut->lut_tdtd; + __u64 transno = 0; + __u64 update_transno; + ENTRY; + + transno = get_next_replay_req_transno(obd); + if (type != NULL) + *type = REQUEST_RECOVERY; + + if (tdtd == NULL) + RETURN(transno); + + update_transno = distribute_txn_get_next_transno(tdtd); + if (transno == 0 || (transno >= update_transno && + update_transno != 0)) { + transno = update_transno; + if (type != NULL) + *type = UPDATE_RECOVERY; + } + + RETURN(transno); +} + +/** + * drop duplicate replay request + * + * Because the operation has been replayed by update recovery, the request + * with the same transno will be dropped and also notify the client to send + * next replay request. + * + * \param[in] env execution environment + * \param[in] obd failover obd device + * \param[in] req request to be dropped + */ +static void drop_duplicate_replay_req(struct lu_env *env, + struct obd_device *obd, + struct ptlrpc_request *req) +{ + DEBUG_REQ(D_HA, req, "remove t"LPD64" from %s because of duplicate" + " update records are found.\n", + lustre_msg_get_transno(req->rq_reqmsg), + libcfs_nid2str(req->rq_peer.nid)); + + /* Right now, only for MDS reint operation update replay and + * normal request replay can have the same transno */ + if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT) { + req_capsule_set(&req->rq_pill, &RQF_MDS_REINT); + req->rq_status = req_capsule_server_pack(&req->rq_pill); + if (likely(req->rq_export)) + target_committed_to_req(req); + lustre_msg_set_transno(req->rq_repmsg, req->rq_transno); + target_send_reply(req, req->rq_status, 0); + } else { + DEBUG_REQ(D_ERROR, req, "wrong opc" "from %s\n", + libcfs_nid2str(req->rq_peer.nid)); + } + target_exp_dequeue_req_replay(req); + target_request_copy_put(req); + obd->obd_replayed_requests++; +} + +static void replay_request_or_update(struct lu_env *env, + struct lu_target *lut, + struct target_recovery_data *trd, + struct ptlrpc_thread *thread) +{ + struct obd_device *obd = lut->lut_obd; + struct ptlrpc_request *req = NULL; + int type; + __u64 transno; + ENTRY; + + CDEBUG(D_HA, "Waiting for transno "LPD64"\n", + obd->obd_next_recovery_transno); + + /* Replay all of request and update by transno */ + do { + struct target_distribute_txn_data *tdtd = lut->lut_tdtd; + + CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLAY_DELAY2, cfs_fail_val); + + /** It is needed to extend recovery window above + * recovery_time_soft. Extending is possible only in the + * end of recovery window (see more details in + * handle_recovery_req()). + */ + CFS_FAIL_TIMEOUT_MS(OBD_FAIL_TGT_REPLAY_DELAY, 300); + + if (target_recovery_overseer(lut, check_for_next_transno, + exp_req_replay_healthy)) { + abort_req_replay_queue(obd); + abort_lock_replay_queue(obd); + } + + spin_lock(&obd->obd_recovery_task_lock); + transno = get_next_transno(lut, &type); + if (type == REQUEST_RECOVERY && transno != 0) { + /* Drop replay request from client side, if the + * replay has been executed by update with the + * same transno */ + req = list_entry(obd->obd_req_replay_queue.next, + struct ptlrpc_request, rq_list); + + list_del_init(&req->rq_list); + obd->obd_requests_queued_for_recovery--; + spin_unlock(&obd->obd_recovery_task_lock); + + /* Let's check if the request has been redone by + * update replay */ + if (is_req_replayed_by_update(req)) { + struct distribute_txn_replay_req *dtrq; + + dtrq = distribute_txn_lookup_finish_list(tdtd, + req->rq_xid); + LASSERT(dtrq != NULL); + spin_lock(&tdtd->tdtd_replay_list_lock); + list_del_init(&dtrq->dtrq_list); + spin_unlock(&tdtd->tdtd_replay_list_lock); + dtrq_destroy(dtrq); + + drop_duplicate_replay_req(env, obd, req); + + continue; + } + + LASSERT(trd->trd_processing_task == current_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); + /** + * bz18031: increase next_recovery_transno before + * target_request_copy_put() will drop exp_rpc reference + */ + spin_lock(&obd->obd_recovery_task_lock); + obd->obd_next_recovery_transno++; + spin_unlock(&obd->obd_recovery_task_lock); + target_exp_dequeue_req_replay(req); + target_request_copy_put(req); + obd->obd_replayed_requests++; + } else if (type == UPDATE_RECOVERY && transno != 0) { + struct distribute_txn_replay_req *dtrq; + + spin_unlock(&obd->obd_recovery_task_lock); + + LASSERT(tdtd != NULL); + dtrq = distribute_txn_get_next_req(tdtd); + lu_context_enter(&thread->t_env->le_ctx); + tdtd->tdtd_replay_handler(env, tdtd, dtrq); + lu_context_exit(&thread->t_env->le_ctx); + extend_recovery_timer(obd, obd_timeout, true); + + /* Add it to the replay finish list */ + spin_lock(&tdtd->tdtd_replay_list_lock); + if (dtrq->dtrq_xid != 0) { + CDEBUG(D_HA, "Move x"LPU64" t"LPU64 + " to finish list\n", dtrq->dtrq_xid, + dtrq->dtrq_master_transno); + list_add(&dtrq->dtrq_list, + &tdtd->tdtd_replay_finish_list); + } else { + dtrq_destroy(dtrq); + } + spin_unlock(&tdtd->tdtd_replay_list_lock); + + spin_lock(&obd->obd_recovery_task_lock); + if (transno == obd->obd_next_recovery_transno) + obd->obd_next_recovery_transno++; + else if (transno > obd->obd_next_recovery_transno) + obd->obd_next_recovery_transno = transno + 1; + spin_unlock(&obd->obd_recovery_task_lock); + + } else { + spin_unlock(&obd->obd_recovery_task_lock); + LASSERT(list_empty(&obd->obd_req_replay_queue)); + LASSERT(atomic_read(&obd->obd_req_replay_clients) == 0); + /** evict exports failed VBR */ + class_disconnect_stale_exports(obd, exp_vbr_healthy); + break; + } + } while (1); +} + static int target_recovery_thread(void *arg) { struct lu_target *lut = arg; @@ -2011,43 +2356,28 @@ static int target_recovery_thread(void *arg) spin_unlock(&obd->obd_dev_lock); complete(&trd->trd_starting); - /* first of all, we have to know the first transno to replay */ - if (target_recovery_overseer(obd, check_for_clients, - exp_connect_healthy)) { - abort_req_replay_queue(obd); - abort_lock_replay_queue(obd); - } + /* first of all, we have to know the first transno to replay */ + if (target_recovery_overseer(lut, check_for_recovery_ready, + exp_connect_healthy)) { + abort_req_replay_queue(obd); + abort_lock_replay_queue(obd); + if (lut->lut_tdtd != NULL) + dtrq_list_destroy(lut->lut_tdtd); + } - /* next stage: replay requests */ + /* next stage: replay requests or update */ delta = jiffies; 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); - while ((req = target_next_replay_req(obd))) { - LASSERT(trd->trd_processing_task == current_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); - /** - * bz18031: increase next_recovery_transno before - * target_request_copy_put() will drop exp_rpc reference - */ - spin_lock(&obd->obd_recovery_task_lock); - obd->obd_next_recovery_transno++; - spin_unlock(&obd->obd_recovery_task_lock); - target_exp_dequeue_req_replay(req); - target_request_copy_put(req); - obd->obd_replayed_requests++; - } + replay_request_or_update(env, lut, trd, thread); /** * 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))) { + while ((req = target_next_replay_lock(lut))) { LASSERT(trd->trd_processing_task == current_pid()); DEBUG_REQ(D_HA, req, "processing lock from %s: ", libcfs_nid2str(req->rq_peer.nid)); @@ -2094,7 +2424,7 @@ static int target_recovery_thread(void *arg) libcfs_debug_dumplog(); } - target_finish_recovery(obd); + target_finish_recovery(lut); lu_context_fini(&env->le_ctx); trd->trd_processing_task = 0; @@ -2170,6 +2500,7 @@ 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; + if (obd->obd_max_recoverable_clients == 0) { /** Update server last boot epoch */ tgt_boot_epoch_update(lut); @@ -2189,7 +2520,6 @@ void target_recovery_init(struct lu_target *lut, svc_handler_t handler) } EXPORT_SYMBOL(target_recovery_init); - static int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req) { @@ -2300,8 +2630,13 @@ int target_queue_recovery_request(struct ptlrpc_request *req, CDEBUG(D_HA, "Next recovery transno: "LPU64 ", current: "LPU64", replaying\n", obd->obd_next_recovery_transno, transno); + + /* If the request has been replayed by update replay, then sends this + * request to the recovery thread (replay_request_or_update()), where + * it will be handled */ spin_lock(&obd->obd_recovery_task_lock); - if (transno < obd->obd_next_recovery_transno) { + if (transno < obd->obd_next_recovery_transno && + !is_req_replayed_by_update(req)) { /* Processing the queue right now, don't re-add. */ LASSERT(list_empty(&req->rq_list)); spin_unlock(&obd->obd_recovery_task_lock); @@ -2410,20 +2745,23 @@ int target_pack_pool_reply(struct ptlrpc_request *req) static int target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id) { - if (OBD_FAIL_CHECK_ORSET(fail_id & ~OBD_FAIL_ONCE, OBD_FAIL_ONCE)) { - DEBUG_REQ(D_ERROR, req, "dropping reply"); - return (-ECOMM); - } + if (OBD_FAIL_CHECK_ORSET(fail_id & ~OBD_FAIL_ONCE, OBD_FAIL_ONCE)) { + DEBUG_REQ(D_ERROR, req, "dropping reply"); + return -ECOMM; + } + if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT && + OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET_REP))) + return -ECOMM; - if (unlikely(rc)) { - DEBUG_REQ(D_NET, req, "processing error (%d)", rc); - req->rq_status = rc; - return (ptlrpc_send_error(req, 1)); - } else { - DEBUG_REQ(D_NET, req, "sending reply"); - } + if (unlikely(rc)) { + DEBUG_REQ(D_NET, req, "processing error (%d)", rc); + req->rq_status = rc; + return ptlrpc_send_error(req, 1); + } else { + DEBUG_REQ(D_NET, req, "sending reply"); + } - return (ptlrpc_send_reply(req, PTLRPC_REPLY_MAYBE_DIFFICULT)); + return ptlrpc_send_reply(req, PTLRPC_REPLY_MAYBE_DIFFICULT); } void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) @@ -2518,7 +2856,7 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) EXIT; } -ldlm_mode_t lck_compat_array[] = { +enum ldlm_mode lck_compat_array[] = { [LCK_EX] = LCK_COMPAT_EX, [LCK_PW] = LCK_COMPAT_PW, [LCK_PR] = LCK_COMPAT_PR, @@ -2533,12 +2871,12 @@ ldlm_mode_t lck_compat_array[] = { * 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 ldlm_error2errno(enum ldlm_error error) { - int result; + int result; - switch (error) { - case ELDLM_OK: + switch (error) { + case ELDLM_OK: case ELDLM_LOCK_MATCHED: result = 0; break; @@ -2560,22 +2898,22 @@ int ldlm_error2errno(ldlm_error_t error) 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: %d\n", error); - result = -EPROTO; - } - } - return result; + default: + if (((int)error) < 0) { /* cast to signed type */ + result = error; /* as ldlm_error can be unsigned */ + } else { + CERROR("Invalid DLM result code: %d\n", error); + result = -EPROTO; + } + } + return result; } EXPORT_SYMBOL(ldlm_error2errno); /** - * Dual to ldlm_error2errno(): maps errno values back to ldlm_error_t. + * Dual to ldlm_error2errno(): maps errno values back to enum ldlm_error. */ -ldlm_error_t ldlm_errno2error(int err_no) +enum ldlm_error ldlm_errno2error(int err_no) { int error; @@ -2669,6 +3007,13 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, } else { if (req->rq_bulk_read) rc = sptlrpc_svc_wrap_bulk(req, desc); + + if ((exp->exp_connect_data.ocd_connect_flags & + OBD_CONNECT_BULK_MBITS) != 0) + req->rq_mbits = lustre_msg_get_mbits(req->rq_reqmsg); + else /* old version, bulk matchbits is rq_xid */ + req->rq_mbits = req->rq_xid; + if (rc == 0) rc = ptlrpc_start_bulk_transfer(desc); }