/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * 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. * * 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). * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ /* * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * * Copyright (c) 2010, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ * Lustre is a trademark of Sun Microsystems, Inc. */ /** * This file deals with various client/target related logic including recovery. * * TODO: This code more logically belongs in the ptlrpc module than in ldlm and * should be moved. */ #define DEBUG_SUBSYSTEM S_LDLM #include #include #include #include #include #include #include #include #include #include #include "ldlm_internal.h" /* * @priority: If non-zero, move the selected connection to the list head. * @create: If zero, only search in existing connections. */ static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid, int priority, int create) { struct ptlrpc_connection *ptlrpc_conn; struct obd_import_conn *imp_conn = NULL, *item; lnet_nid_t nid4refnet = LNET_NID_ANY; int rc = 0; ENTRY; if (!create && !priority) { CDEBUG(D_HA, "Nothing to do\n"); RETURN(-EINVAL); } if (imp->imp_connection && imp->imp_connection->c_remote_uuid.uuid[0] == 0) /* nid4refnet is used to restrict network connections */ nid4refnet = imp->imp_connection->c_self; ptlrpc_conn = ptlrpc_uuid_to_connection(uuid, nid4refnet); if (!ptlrpc_conn) { CDEBUG(D_HA, "can't find connection %s\n", uuid->uuid); RETURN(-ENOENT); } if (create) { OBD_ALLOC(imp_conn, sizeof(*imp_conn)); if (!imp_conn) GOTO(out_put, rc = -ENOMEM); } spin_lock(&imp->imp_lock); list_for_each_entry(item, &imp->imp_conn_list, oic_item) { if (obd_uuid_equals(uuid, &item->oic_uuid)) { if (priority) { list_move(&item->oic_item, &imp->imp_conn_list); item->oic_last_attempt = 0; } CDEBUG(D_HA, "imp %p@%s: found existing conn %s%s\n", imp, imp->imp_obd->obd_name, uuid->uuid, (priority ? ", moved to head" : "")); spin_unlock(&imp->imp_lock); GOTO(out_free, rc = 0); } } /* No existing import connection found for \a uuid. */ if (create) { imp_conn->oic_conn = ptlrpc_conn; imp_conn->oic_uuid = *uuid; imp_conn->oic_last_attempt = 0; if (priority) list_add(&imp_conn->oic_item, &imp->imp_conn_list); else list_add_tail(&imp_conn->oic_item, &imp->imp_conn_list); CDEBUG(D_HA, "imp %p@%s: add connection %s at %s\n", imp, imp->imp_obd->obd_name, uuid->uuid, (priority ? "head" : "tail")); } else { spin_unlock(&imp->imp_lock); GOTO(out_free, rc = -ENOENT); } spin_unlock(&imp->imp_lock); RETURN(0); out_free: if (imp_conn) OBD_FREE(imp_conn, sizeof(*imp_conn)); out_put: ptlrpc_connection_put(ptlrpc_conn); RETURN(rc); } int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid) { return import_set_conn(imp, uuid, 1, 0); } int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid, int priority) { return import_set_conn(imp, uuid, priority, 1); } EXPORT_SYMBOL(client_import_add_conn); int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid) { struct obd_import_conn *imp_conn; struct obd_export *dlmexp; int rc = -ENOENT; ENTRY; spin_lock(&imp->imp_lock); if (list_empty(&imp->imp_conn_list)) { LASSERT(!imp->imp_connection); GOTO(out, rc); } list_for_each_entry(imp_conn, &imp->imp_conn_list, oic_item) { if (!obd_uuid_equals(uuid, &imp_conn->oic_uuid)) continue; LASSERT(imp_conn->oic_conn); if (imp_conn == imp->imp_conn_current) { LASSERT(imp_conn->oic_conn == imp->imp_connection); if (imp->imp_state != LUSTRE_IMP_CLOSED && imp->imp_state != LUSTRE_IMP_DISCON) { CERROR("can't remove current connection\n"); GOTO(out, rc = -EBUSY); } 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_connection_put(dlmexp->exp_connection); dlmexp->exp_connection = NULL; } if (dlmexp != NULL) class_export_put(dlmexp); } list_del(&imp_conn->oic_item); 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); rc = 0; break; } out: spin_unlock(&imp->imp_lock); if (rc == -ENOENT) CERROR("connection %s not found\n", uuid->uuid); RETURN(rc); } EXPORT_SYMBOL(client_import_del_conn); /** * Find conn UUID by peer NID. \a peer is a server NID. This function is used * to find a conn uuid of \a imp which can reach \a peer. */ int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer, struct obd_uuid *uuid) { struct obd_import_conn *conn; int rc = -ENOENT; ENTRY; spin_lock(&imp->imp_lock); list_for_each_entry(conn, &imp->imp_conn_list, oic_item) { /* Check if conn UUID does have this peer NID. */ if (class_check_uuid(&conn->oic_uuid, peer)) { *uuid = conn->oic_uuid; rc = 0; break; } } spin_unlock(&imp->imp_lock); RETURN(rc); } EXPORT_SYMBOL(client_import_find_conn); void client_destroy_import(struct obd_import *imp) { /* * Drop security policy instance after all RPCs have finished/aborted * to let all busy contexts be released. */ class_import_get(imp); class_destroy_import(imp); sptlrpc_import_sec_put(imp); class_import_put(imp); } EXPORT_SYMBOL(client_destroy_import); /** * Check whether or not the OSC is on MDT. * In the config log, * osc on MDT * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID * osc on client * setup 0:{fsname}-OSTxxxx-osc 1:lustre-OST0000_UUID 2:NID * **/ static int osc_on_mdt(char *obdname) { char *ptr; ptr = strrchr(obdname, '-'); if (ptr == NULL) return 0; if (strncmp(ptr + 1, "MDT", 3) == 0) return 1; return 0; } /* * Configure an RPC client OBD device. * * lcfg parameters: * 1 - client UUID * 2 - server UUID * 3 - inactive-on-startup * 4 - restrictive net */ int client_obd_setup(struct obd_device *obd, struct lustre_cfg *lcfg) { struct client_obd *cli = &obd->u.cli; struct obd_import *imp; struct obd_uuid server_uuid; int rq_portal, rp_portal, connect_op; const char *name = obd->obd_type->typ_name; enum ldlm_ns_type ns_type = LDLM_NS_TYPE_UNKNOWN; char *cli_name = lustre_cfg_buf(lcfg, 0); struct ptlrpc_connection fake_conn = { .c_self = 0, .c_remote_uuid.uuid[0] = 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. */ if (!strcmp(name, LUSTRE_OSC_NAME)) { 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; ns_type = LDLM_NS_TYPE_OSC; } else if (!strcmp(name, LUSTRE_MDC_NAME) || !strcmp(name, LUSTRE_LWP_NAME)) { rq_portal = MDS_REQUEST_PORTAL; rp_portal = MDC_REPLY_PORTAL; connect_op = MDS_CONNECT; 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)) { if (strstr(lustre_cfg_buf(lcfg, 1), "OST") == NULL) { /* OSP_on_MDT for other MDTs */ connect_op = MDS_CONNECT; cli->cl_sp_to = LUSTRE_SP_MDT; ns_type = LDLM_NS_TYPE_MDC; rq_portal = OUT_PORTAL; } else { /* OSP on MDT for OST */ connect_op = OST_CONNECT; cli->cl_sp_to = LUSTRE_SP_OST; ns_type = LDLM_NS_TYPE_OSC; rq_portal = OST_REQUEST_PORTAL; } rp_portal = OSC_REPLY_PORTAL; cli->cl_sp_me = 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; ns_type = LDLM_NS_TYPE_MGC; } else { CERROR("unknown client OBD type \"%s\", can't setup\n", name); RETURN(-EINVAL); } if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) { CERROR("requires a TARGET UUID\n"); RETURN(-EINVAL); } if (LUSTRE_CFG_BUFLEN(lcfg, 1) > 37) { CERROR("client UUID must be less than 38 characters\n"); RETURN(-EINVAL); } if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) { CERROR("setup requires a SERVER UUID\n"); RETURN(-EINVAL); } if (LUSTRE_CFG_BUFLEN(lcfg, 2) > 37) { CERROR("target UUID must be less than 38 characters\n"); RETURN(-EINVAL); } init_rwsem(&cli->cl_sem); mutex_init(&cli->cl_mgc_mutex); cli->cl_seq = NULL; init_rwsem(&cli->cl_seq_rwsem); cli->cl_conn_count = 0; memcpy(server_uuid.uuid, lustre_cfg_buf(lcfg, 2), min_t(unsigned int, LUSTRE_CFG_BUFLEN(lcfg, 2), sizeof(server_uuid))); cli->cl_dirty_pages = 0; cli->cl_dirty_max_pages = 0; cli->cl_avail_grant = 0; /* FIXME: Should limit this for the sum of all cl_dirty_max_pages. */ /* * cl_dirty_max_pages may be changed at connect time in * ptlrpc_connect_interpret(). */ client_adjust_max_dirty(cli); INIT_LIST_HEAD(&cli->cl_cache_waiters); INIT_LIST_HEAD(&cli->cl_loi_ready_list); INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list); INIT_LIST_HEAD(&cli->cl_loi_write_list); INIT_LIST_HEAD(&cli->cl_loi_read_list); spin_lock_init(&cli->cl_loi_list_lock); atomic_set(&cli->cl_pending_w_pages, 0); atomic_set(&cli->cl_pending_r_pages, 0); cli->cl_r_in_flight = 0; cli->cl_w_in_flight = 0; spin_lock_init(&cli->cl_read_rpc_hist.oh_lock); spin_lock_init(&cli->cl_write_rpc_hist.oh_lock); spin_lock_init(&cli->cl_read_page_hist.oh_lock); spin_lock_init(&cli->cl_write_page_hist.oh_lock); spin_lock_init(&cli->cl_read_offset_hist.oh_lock); spin_lock_init(&cli->cl_write_offset_hist.oh_lock); /* lru for osc. */ INIT_LIST_HEAD(&cli->cl_lru_osc); atomic_set(&cli->cl_lru_shrinkers, 0); atomic_long_set(&cli->cl_lru_busy, 0); atomic_long_set(&cli->cl_lru_in_list, 0); INIT_LIST_HEAD(&cli->cl_lru_list); spin_lock_init(&cli->cl_lru_list_lock); atomic_long_set(&cli->cl_unstable_count, 0); INIT_LIST_HEAD(&cli->cl_shrink_list); INIT_LIST_HEAD(&cli->cl_grant_chain); INIT_LIST_HEAD(&cli->cl_flight_waiters); cli->cl_rpcs_in_flight = 0; init_waitqueue_head(&cli->cl_destroy_waitq); atomic_set(&cli->cl_destroy_in_flight, 0); cli->cl_supp_cksum_types = OBD_CKSUM_CRC32; cli->cl_preferred_cksum_type = 0; #ifdef ENABLE_CHECKSUM /* Turn on checksumming by default. */ cli->cl_checksum = 1; /* * The supported checksum types will be worked out at connect time * Set cl_chksum* to CRC32 for now to avoid returning screwed info * through procfs. */ cli->cl_cksum_type = cli->cl_supp_cksum_types; #endif atomic_set(&cli->cl_resends, OSC_DEFAULT_RESENDS); /* * Set it to possible maximum size. It may be reduced by ocd_brw_size * from OFD after connecting. */ cli->cl_max_pages_per_rpc = PTLRPC_MAX_BRW_PAGES; cli->cl_max_short_io_bytes = OBD_DEF_SHORT_IO_BYTES; /* * set cl_chunkbits default value to PAGE_SHIFT, * it will be updated at OSC connection time. */ cli->cl_chunkbits = PAGE_SHIFT; if (!strcmp(name, LUSTRE_MDC_NAME)) { cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_DEFAULT; } else if (cfs_totalram_pages() >> (20 - PAGE_SHIFT) <= 128 /* MB */) { cli->cl_max_rpcs_in_flight = 2; } else if (cfs_totalram_pages() >> (20 - PAGE_SHIFT) <= 256 /* MB */) { cli->cl_max_rpcs_in_flight = 3; } else if (cfs_totalram_pages() >> (20 - PAGE_SHIFT) <= 512 /* MB */) { cli->cl_max_rpcs_in_flight = 4; } else { if (osc_on_mdt(obd->obd_name)) cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_MAX; else cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_DEFAULT; } 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; INIT_LIST_HEAD(&cli->cl_chg_dev_linkage); if (connect_op == MDS_CONNECT) { cli->cl_max_mod_rpcs_in_flight = cli->cl_max_rpcs_in_flight - 1; OBD_ALLOC(cli->cl_mod_tag_bitmap, 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); GOTO(err, rc); } ptlrpc_init_client(rq_portal, rp_portal, name, &obd->obd_ldlm_client); imp = class_new_import(obd); if (imp == NULL) GOTO(err_ldlm, rc = -ENOENT); imp->imp_client = &obd->obd_ldlm_client; imp->imp_connect_op = connect_op; memcpy(cli->cl_target_uuid.uuid, lustre_cfg_buf(lcfg, 1), LUSTRE_CFG_BUFLEN(lcfg, 1)); class_import_put(imp); if (lustre_cfg_buf(lcfg, 4)) { __u32 refnet = libcfs_str2net(lustre_cfg_string(lcfg, 4)); if (refnet == LNET_NIDNET(LNET_NID_ANY)) { rc = -EINVAL; CERROR("%s: bad mount option 'network=%s': rc = %d\n", obd->obd_name, lustre_cfg_string(lcfg, 4), rc); GOTO(err_import, rc); } fake_conn.c_self = LNET_MKNID(refnet, 0); imp->imp_connection = &fake_conn; } rc = client_import_add_conn(imp, &server_uuid, 1); if (rc) { CERROR("can't add initial connection\n"); GOTO(err_import, rc); } imp->imp_connection = NULL; cli->cl_import = imp; /* cli->cl_max_mds_easize updated by mdc_init_ea_size() */ 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")) { CDEBUG(D_HA, "marking %s %s->%s as inactive\n", name, obd->obd_name, cli->cl_target_uuid.uuid); spin_lock(&imp->imp_lock); imp->imp_deactive = 1; spin_unlock(&imp->imp_lock); } } obd->obd_namespace = ldlm_namespace_new(obd, obd->obd_name, LDLM_NAMESPACE_CLIENT, LDLM_NAMESPACE_GREEDY, ns_type); if (obd->obd_namespace == NULL) { CERROR("Unable to create client namespace - %s\n", obd->obd_name); GOTO(err_import, rc = -ENOMEM); } RETURN(rc); err_import: class_destroy_import(imp); 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); } EXPORT_SYMBOL(client_obd_setup); int client_obd_cleanup(struct obd_device *obd) { struct client_obd *cli = &obd->u.cli; ENTRY; ldlm_namespace_free_post(obd->obd_namespace); obd->obd_namespace = NULL; obd_cleanup_client_import(obd); LASSERT(obd->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); /* ->o_connect() method for client side (OSC and MDC and MGC) */ int client_connect_import(const struct lu_env *env, struct obd_export **exp, struct obd_device *obd, struct obd_uuid *cluuid, struct obd_connect_data *data, void *localdata) { struct client_obd *cli = &obd->u.cli; struct obd_import *imp = cli->cl_import; struct obd_connect_data *ocd; struct lustre_handle conn = { 0 }; int rc; ENTRY; *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++; *exp = class_conn2export(&conn); LASSERT(obd->obd_namespace); imp->imp_dlm_handle = conn; rc = ptlrpc_init_import(imp); if (rc != 0) GOTO(out_ldlm, rc); ocd = &imp->imp_connect_data; if (data) { *ocd = *data; imp->imp_connect_flags_orig = data->ocd_connect_flags; imp->imp_connect_flags2_orig = data->ocd_connect_flags2; } rc = ptlrpc_connect_import(imp); if (rc != 0) { LASSERT(imp->imp_state == LUSTRE_IMP_DISCON); GOTO(out_ldlm, rc); } LASSERT(*exp != NULL && (*exp)->exp_connection); if (data) { LASSERTF((ocd->ocd_connect_flags & data->ocd_connect_flags) == ocd->ocd_connect_flags, "old %#llx, new %#llx\n", data->ocd_connect_flags, ocd->ocd_connect_flags); data->ocd_connect_flags = ocd->ocd_connect_flags; data->ocd_connect_flags2 = ocd->ocd_connect_flags2; } ptlrpc_pinger_add_import(imp); EXIT; if (rc) { out_ldlm: cli->cl_conn_count--; class_disconnect(*exp); *exp = NULL; } out_sem: up_write(&cli->cl_sem); if (!rc && localdata) { LASSERT(cli->cl_cache == NULL); /* only once */ cli->cl_cache = (struct cl_client_cache *)localdata; cl_cache_incref(cli->cl_cache); cli->cl_lru_left = &cli->cl_cache->ccc_lru_left; /* add this osc into entity list */ LASSERT(list_empty(&cli->cl_lru_osc)); spin_lock(&cli->cl_cache->ccc_lru_lock); list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru); spin_unlock(&cli->cl_cache->ccc_lru_lock); } return rc; } EXPORT_SYMBOL(client_connect_import); int client_disconnect_export(struct obd_export *exp) { struct obd_device *obd = class_exp2obd(exp); struct client_obd *cli; struct obd_import *imp; int rc = 0, err; ENTRY; if (!obd) { CERROR("invalid export for disconnect: exp %p cookie %#llx\n", exp, exp ? exp->exp_handle.h_cookie : -1); RETURN(-EINVAL); } cli = &obd->u.cli; imp = cli->cl_import; down_write(&cli->cl_sem); CDEBUG(D_INFO, "disconnect %s - %zu\n", obd->obd_name, cli->cl_conn_count); if (cli->cl_conn_count == 0) { CERROR("disconnecting disconnected device (%s)\n", obd->obd_name); GOTO(out_disconnect, rc = -EINVAL); } cli->cl_conn_count--; if (cli->cl_conn_count != 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 * fully deactivate the import, or that would drop all requests. */ spin_lock(&imp->imp_lock); imp->imp_deactive = 1; spin_unlock(&imp->imp_lock); /* * Some non-replayable imports (MDS's OSCs) are pinged, so just * delete it regardless. (It's safe to delete an import that was * never added.) */ (void)ptlrpc_pinger_del_import(imp); if (obd->obd_namespace != NULL) { /* obd_force == local only */ ldlm_cli_cancel_unused(obd->obd_namespace, NULL, obd->obd_force ? LCF_LOCAL : 0, NULL); ldlm_namespace_free_prior(obd->obd_namespace, imp, obd->obd_force); } /* * There's no need to hold sem while disconnecting an import, * and it may actually cause deadlock in GSS. */ up_write(&cli->cl_sem); rc = ptlrpc_disconnect_import(imp, 0); down_write(&cli->cl_sem); ptlrpc_invalidate_import(imp); EXIT; out_disconnect: /* * Use server style - class_disconnect should be always called for * o_disconnect. */ err = class_disconnect(exp); if (!rc && err) rc = err; up_write(&cli->cl_sem); RETURN(rc); } EXPORT_SYMBOL(client_disconnect_export); #ifdef HAVE_SERVER_SUPPORT int server_disconnect_export(struct obd_export *exp) { int rc; ENTRY; /* Disconnect early so that clients can't keep using export. */ rc = class_disconnect(exp); /* Close import to avoid sending any requests. */ if (exp->exp_imp_reverse) ptlrpc_cleanup_imp(exp->exp_imp_reverse); ldlm_bl_thread_wakeup(); /* complete all outstanding replies */ spin_lock(&exp->exp_lock); while (!list_empty(&exp->exp_outstanding_replies)) { struct ptlrpc_reply_state *rs = list_entry(exp->exp_outstanding_replies.next, struct ptlrpc_reply_state, rs_exp_list); struct ptlrpc_service_part *svcpt = rs->rs_svcpt; spin_lock(&svcpt->scp_rep_lock); list_del_init(&rs->rs_exp_list); spin_lock(&rs->rs_lock); /* clear rs_convert_lock to make sure rs is handled and put */ rs->rs_convert_lock = 0; ptlrpc_schedule_difficult_reply(rs); spin_unlock(&rs->rs_lock); spin_unlock(&svcpt->scp_rep_lock); } spin_unlock(&exp->exp_lock); RETURN(rc); } EXPORT_SYMBOL(server_disconnect_export); /* * -------------------------------------------------------------------------- * from old lib/target.c * -------------------------------------------------------------------------- */ static int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp, struct obd_uuid *cluuid) { struct obd_device *target; struct lustre_handle *hdl; ktime_t remaining; s64 timeout; int rc = 0; ENTRY; hdl = &exp->exp_imp_reverse->imp_remote_handle; if (!exp->exp_connection || !lustre_handle_is_used(hdl)) { conn->cookie = exp->exp_handle.h_cookie; CDEBUG(D_HA, "connect export for UUID '%s' at %p, cookie %#llx\n", cluuid->uuid, exp, conn->cookie); RETURN(0); } target = exp->exp_obd; /* Might be a re-connect after a partition. */ if (memcmp(&conn->cookie, &hdl->cookie, sizeof(conn->cookie))) { LCONSOLE_WARN("%s: already connected client %s (at %s) with handle %#llx. Rejecting client with the same UUID trying to reconnect with handle %#llx\n", target->obd_name, obd_uuid2str(&exp->exp_client_uuid), obd_export_nid2str(exp), hdl->cookie, conn->cookie); memset(conn, 0, sizeof(*conn)); /* * target_handle_connect() treats EALREADY and * -EALREADY differently. -EALREADY is an error * (same UUID, different handle). */ RETURN(-EALREADY); } if (!target->obd_recovering) { LCONSOLE_WARN("%s: Client %s (at %s) reconnecting\n", target->obd_name, obd_uuid2str(&exp->exp_client_uuid), obd_export_nid2str(exp)); GOTO(out_already, rc); } remaining = hrtimer_expires_remaining(&target->obd_recovery_timer); timeout = ktime_divns(remaining, NSEC_PER_SEC); if (timeout > 0) { LCONSOLE_WARN("%s: Client %s (at %s) reconnected, waiting for %d clients in recovery for %lld:%.02lld\n", target->obd_name, obd_uuid2str(&exp->exp_client_uuid), obd_export_nid2str(exp), atomic_read(&target->obd_max_recoverable_clients), timeout / 60, timeout % 60); } else { struct target_distribute_txn_data *tdtd; int size = 0; int count = 0; char *buf = NULL; tdtd = class_exp2tgt(exp)->lut_tdtd; if (tdtd && tdtd->tdtd_show_update_logs_retrievers) buf = tdtd->tdtd_show_update_logs_retrievers( tdtd->tdtd_show_retrievers_cbdata, &size, &count); if (count > 0) LCONSOLE_WARN("%s: Client %s (at %s) reconnecting, waiting for %d MDTs (%s) in recovery for %lld:%.02lld. Please wait until all MDTs recovered or you may force MDT evicition via 'lctl --device %s abort_recovery.\n", target->obd_name, obd_uuid2str(&exp->exp_client_uuid), obd_export_nid2str(exp), count, buf ? buf : "unknown (not enough RAM)", (abs(timeout) + target->obd_recovery_timeout) / 60, (abs(timeout) + target->obd_recovery_timeout) % 60, target->obd_name); else LCONSOLE_WARN("%s: Recovery already passed deadline %lld:%.02lld. If you do not want to wait more, you may force taget eviction via 'lctl --device %s abort_recovery.\n", target->obd_name, abs(timeout) / 60, abs(timeout) % 60, target->obd_name); if (buf != NULL) OBD_FREE(buf, size); } out_already: conn->cookie = exp->exp_handle.h_cookie; /* * target_handle_connect() treats EALREADY and * -EALREADY differently. EALREADY means we are * doing a valid reconnect from the same client. */ RETURN(EALREADY); } 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; 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 cluuid; char *str; int rc = 0; char *target_start; int target_len; bool mds_conn = false, lw_client = false, initial_conn = 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; ENTRY; OBD_RACE(OBD_FAIL_TGT_CONN_RACE); str = req_capsule_client_get(&req->rq_pill, &RMF_TGTUUID); if (str == NULL) { DEBUG_REQ(D_ERROR, req, "bad target UUID for connect"); GOTO(out, rc = -EINVAL); } target = class_dev_by_str(str); if (!target) { deuuidify(str, NULL, &target_start, &target_len); LCONSOLE_ERROR_MSG(0x137, "%s: not available for connect from %s (no target). If you are running an HA pair check that the target is mounted on the other server.\n", str, libcfs_nid2str(req->rq_peer.nid)); GOTO(out, rc = -ENODEV); } spin_lock(&target->obd_dev_lock); target->obd_conn_inprogress++; if (target->obd_stopping || !target->obd_set_up) { spin_unlock(&target->obd_dev_lock); deuuidify(str, NULL, &target_start, &target_len); LCONSOLE_INFO("%.*s: Not available for connect from %s (%s)\n", target_len, target_start, libcfs_nid2str(req->rq_peer.nid), (target->obd_stopping ? "stopping" : "not set up")); GOTO(out, rc = -ENODEV); } if (target->obd_no_conn) { spin_unlock(&target->obd_dev_lock); CDEBUG(D_INFO, "%s: Temporarily refusing client connection from %s\n", target->obd_name, libcfs_nid2str(req->rq_peer.nid)); GOTO(out, rc = -EAGAIN); } spin_unlock(&target->obd_dev_lock); str = req_capsule_client_get(&req->rq_pill, &RMF_CLUUID); if (str == NULL) { DEBUG_REQ(D_ERROR, req, "bad client UUID for connect"); GOTO(out, rc = -EINVAL); } obd_str2uuid(&cluuid, str); tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN); if (tmp == NULL) GOTO(out, rc = -EPROTO); conn = *tmp; size = req_capsule_get_size(&req->rq_pill, &RMF_CONNECT_DATA, RCL_CLIENT); if (size < 0 || size > 8 * sizeof(struct obd_connect_data)) GOTO(out, rc = -EPROTO); data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA); if (!data) GOTO(out, rc = -EPROTO); rc = req_capsule_server_pack(&req->rq_pill); 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); /* * Don't allow liblustre clients to connect. * - testing was disabled in v2_2_50_0-61-g6a75d65 * - building was disabled in v2_5_58_0-28-g7277179 * - client code was deleted in v2_6_50_0-101-gcdfbc72, * - clients were refused connect for version difference > 0.0.1.32 */ if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) { DEBUG_REQ(D_WARNING, req, "Refusing libclient connection"); GOTO(out, rc = -EPROTO); } #endif /* * Note: lw_client is needed in MDS-MDS failover during update log * processing, so we needs to allow lw_client to be connected at * 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) { initial_conn = true; mds_conn = (data->ocd_connect_flags & OBD_CONNECT_MDS) != 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. * * Via check OBD_CONNECT_FID, we can distinguish whether * the OBD_CONNECT_MDS_MDS/OBD_CONNECT_MNE_SWAB is from * MGC or MDT, since MGC does not use OBD_CONNECT_FID. */ if (!lw_client && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) && (data->ocd_connect_flags & OBD_CONNECT_FID) && (data->ocd_connect_flags & OBD_CONNECT_VERSION)) { __u32 major = OBD_OCD_VERSION_MAJOR(data->ocd_version); __u32 minor = OBD_OCD_VERSION_MINOR(data->ocd_version); __u32 patch = OBD_OCD_VERSION_PATCH(data->ocd_version); /* * We do not support the MDT-MDT interoperations with * different version MDT because of protocol changes. */ if (unlikely(major != LUSTRE_MAJOR || minor != LUSTRE_MINOR || abs(patch - LUSTRE_PATCH) > 3)) { LCONSOLE_WARN("%s (%u.%u.%u.%u) refused the connection from different version MDT (%d.%d.%d.%d) %s %s\n", target->obd_name, LUSTRE_MAJOR, LUSTRE_MINOR, LUSTRE_PATCH, LUSTRE_FIX, major, minor, patch, OBD_OCD_VERSION_FIX(data->ocd_version), libcfs_nid2str(req->rq_peer.nid), str); GOTO(out, rc = -EPROTO); } } } /* lctl gets a backstage, all-access pass. */ if (obd_uuid_equals(&cluuid, &target->obd_uuid)) goto dont_check_exports; export = obd_uuid_lookup(target, &cluuid); if (!export) goto no_export; /* We've found an export in the hash. */ spin_lock(&export->exp_lock); if (export->exp_connecting) { /* b=9635, et. al. */ spin_unlock(&export->exp_lock); LCONSOLE_WARN("%s: Export %p already connecting from %s\n", export->exp_obd->obd_name, export, libcfs_nid2str(req->rq_peer.nid)); class_export_put(export); export = NULL; rc = -EALREADY; } else if ((mds_conn || (lw_client && initial_conn) || data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) && export->exp_connection != NULL) { spin_unlock(&export->exp_lock); if (req->rq_peer.nid != export->exp_connection->c_peer.nid) { /* MDS or LWP reconnected after failover. */ LCONSOLE_WARN("%s: Received %s connection from %s, removing former export from %s\n", target->obd_name, mds_conn ? "MDS" : "LWP", libcfs_nid2str(req->rq_peer.nid), libcfs_nid2str(export->exp_connection->c_peer.nid)); } else { /* New MDS connection from the same NID. */ LCONSOLE_WARN("%s: Received new %s connection from %s, removing former export from same NID\n", target->obd_name, mds_conn ? "MDS" : "LWP", libcfs_nid2str(req->rq_peer.nid)); } if (req->rq_peer.nid == export->exp_connection->c_peer.nid && data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) { /* * Because exports between MDTs will always be * kept, let's do not fail such export if they * come from the same NID, otherwise it might * cause eviction between MDTs, which might * cause namespace inconsistency */ spin_lock(&export->exp_lock); export->exp_connecting = 1; export->exp_conn_cnt = 0; spin_unlock(&export->exp_lock); conn.cookie = export->exp_handle.h_cookie; rc = EALREADY; } else { class_fail_export(export); class_export_put(export); export = NULL; rc = 0; } } else if (export->exp_connection != NULL && initial_conn && req->rq_peer.nid != export->exp_connection->c_peer.nid) { spin_unlock(&export->exp_lock); /* In MDS failover we have static UUID but NID can change. */ LCONSOLE_WARN("%s: Client %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 if (OBD_FAIL_PRECHECK(OBD_FAIL_TGT_RECOVERY_CONNECT) && !lw_client) { spin_unlock(&export->exp_lock); rc = -EAGAIN; } else { export->exp_connecting = 1; spin_unlock(&export->exp_lock); LASSERT(export->exp_obd == target); rc = target_handle_reconnect(&conn, export, &cluuid); } /* If we found an export, we already unlocked. */ if (!export) { 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) { LCONSOLE_WARN("%s: Client %s (at %s) refused connection, still busy with %d references\n", target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), refcount_read(&export->exp_handle.h_ref)); GOTO(out, rc = -EBUSY); } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1 && rc != EALREADY) { if (!strstr(cluuid.uuid, "mdt")) LCONSOLE_WARN("%s: Rejecting reconnect from the known client %s (at %s) because it is indicating it is a new client", target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid)); GOTO(out, rc = -EALREADY); } else { OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_RECONNECT, 2 * obd_timeout); } if (rc < 0) GOTO(out, rc); CDEBUG(D_HA, "%s: connection from %s@%s %st%llu exp %p cur %lld last %lld\n", target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), target->obd_recovering ? "recovering/" : "", data->ocd_transno, export, ktime_get_seconds(), export ? export->exp_last_request_time : 0); /* * If this is the first time a client connects, reset the recovery * timer. Discard lightweight connections which might be local. */ if (!lw_client && rc == 0 && target->obd_recovering) check_and_start_recovery_timer(target, req, export == NULL); /* * 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) { /* allow lightweight connections during recovery */ /* * 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) { struct hrtimer *timer = &target->obd_recovery_timer; ktime_t remaining; s64 timeout, left; int in_progress; int connected; int known; int stale; char *msg; connected = atomic_read(&target->obd_connected_clients); in_progress = atomic_read(&target->obd_lock_replay_clients); known = atomic_read(&target->obd_max_recoverable_clients); stale = target->obd_stale_clients; remaining = hrtimer_expires_remaining(timer); left = ktime_divns(remaining, NSEC_PER_SEC); if (ktime_to_ns(remaining) > 0) { msg = "to recover in"; timeout = left; } else { msg = "already passed deadline"; timeout = -left; } LCONSOLE_WARN("%s: Denying connection for new client %s (at %s), waiting for %d known clients (%d recovered, %d in progress, and %d evicted) %s %lld:%.02lld\n", target->obd_name, cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), known, connected - in_progress, in_progress, stale, msg, timeout / 60, timeout % 60); rc = -EBUSY; } else { dont_check_exports: 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; 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); LASSERT(target->u.obt.obt_magic == OBT_MAGIC); data->ocd_instance = target->u.obt.obt_instance; /* * Return only the parts of obd_connect_data that we understand, so the * client knows that we don't understand the rest. */ if (data) { tmpsize = req_capsule_get_size(&req->rq_pill, &RMF_CONNECT_DATA, RCL_SERVER); tmpdata = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA); /* * Don't use struct assignment here, because the client reply * buffer may be smaller/larger than the local struct * obd_connect_data. */ memcpy(tmpdata, data, min(tmpsize, size)); } /* * 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. * * XXX this will go away when shaver stops sending the "connect" handle * in the real "remote handle" field of the request --phik 24 Apr 2003 */ ptlrpc_request_change_export(req, export); spin_lock(&export->exp_lock); if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) { spin_unlock(&export->exp_lock); CDEBUG(D_RPCTRACE, "%s: %s already connected at greater or equal conn_cnt: %d >= %d\n", cluuid.uuid, libcfs_nid2str(req->rq_peer.nid), export->exp_conn_cnt, lustre_msg_get_conn_cnt(req->rq_reqmsg)); GOTO(out, rc = -EALREADY); } LASSERT(lustre_msg_get_conn_cnt(req->rq_reqmsg) > 0); export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg); spin_unlock(&export->exp_lock); if (export->exp_connection != NULL) { /* Check to see if connection came from another NID. */ if ((export->exp_connection->c_peer.nid != req->rq_peer.nid) && !hlist_unhashed(&export->exp_nid_hash)) cfs_hash_del(export->exp_obd->obd_nid_hash, &export->exp_connection->c_peer.nid, &export->exp_nid_hash); ptlrpc_connection_put(export->exp_connection); } 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; __u64 transno = data->ocd_transno; spin_lock(&export->exp_lock); /* * possible race with class_disconnect_stale_exports, * export may be already in the eviction process */ if (export->exp_failed) { spin_unlock(&export->exp_lock); GOTO(out, rc = -ENODEV); } export->exp_in_recovery = 1; export->exp_req_replay_needed = 1; export->exp_lock_replay_needed = 1; spin_unlock(&export->exp_lock); has_transno = !!(lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_TRANSNO); if (has_transno && transno == 0) CWARN("Connect with zero transno!\n"); if (has_transno && transno > 0 && transno < target->obd_next_recovery_transno && transno > target->obd_last_committed) { /* Another way is to use cmpxchg() to be lock-free. */ spin_lock(&target->obd_recovery_task_lock); if (transno < target->obd_next_recovery_transno) target->obd_next_recovery_transno = transno; spin_unlock(&target->obd_recovery_task_lock); } 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) atomic_inc(&target->obd_max_recoverable_clients); if (atomic_inc_return(&target->obd_connected_clients) == atomic_read(&target->obd_max_recoverable_clients)) wake_up(&target->obd_next_transno_waitq); } /* Tell the client we're in recovery, when client is involved in it. */ if (target->obd_recovering && !lw_client) lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING); out: if (export) { spin_lock(&export->exp_lock); export->exp_connecting = 0; spin_unlock(&export->exp_lock); class_export_put(export); } if (target != NULL) { spin_lock(&target->obd_dev_lock); target->obd_conn_inprogress--; spin_unlock(&target->obd_dev_lock); class_decref(target, "find", current); } req->rq_status = rc; RETURN(rc); } int target_handle_disconnect(struct ptlrpc_request *req) { int rc; ENTRY; rc = req_capsule_server_pack(&req->rq_pill); if (rc) RETURN(rc); /* Keep the rq_export around so we can send the reply. */ req->rq_status = obd_disconnect(class_export_get(req->rq_export)); RETURN(0); } void target_destroy_export(struct obd_export *exp) { struct obd_import *imp = NULL; /* * exports created from last_rcvd data, and "fake" * exports created by lctl don't have an import */ spin_lock(&exp->exp_lock); if (exp->exp_imp_reverse != NULL) { imp = exp->exp_imp_reverse; exp->exp_imp_reverse = NULL; } spin_unlock(&exp->exp_lock); if (imp != NULL) client_destroy_import(imp); LASSERT_ATOMIC_ZERO(&exp->exp_locks_count); LASSERT_ATOMIC_ZERO(&exp->exp_rpc_count); LASSERT_ATOMIC_ZERO(&exp->exp_cb_count); LASSERT_ATOMIC_ZERO(&exp->exp_replay_count); } EXPORT_SYMBOL(target_destroy_export); /* * Recovery functions */ static void target_request_copy_get(struct ptlrpc_request *req) { class_export_rpc_inc(req->rq_export); LASSERT(list_empty(&req->rq_list)); INIT_LIST_HEAD(&req->rq_replay_list); /* Increase refcount to keep request in queue. */ atomic_inc(&req->rq_refcount); /* Let export know it has replays to be handled. */ atomic_inc(&req->rq_export->exp_replay_count); } static void target_request_copy_put(struct ptlrpc_request *req) { LASSERT(list_empty(&req->rq_replay_list)); LASSERT_ATOMIC_POS(&req->rq_export->exp_replay_count); atomic_dec(&req->rq_export->exp_replay_count); class_export_rpc_dec(req->rq_export); ptlrpc_server_drop_request(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; struct ptlrpc_request *dup_req = NULL; 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_req = reqiter; 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)); 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); } 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); } 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. */ if (obd->obd_recovery_start) { time64_t now = ktime_get_seconds(); time64_t elapsed_time; elapsed_time = max_t(time64_t, now - obd->obd_recovery_start, 1); LCONSOLE_INFO("%s: Recovery over after %lld:%.02lld, of %d clients %d recovered and %d %s evicted.\n", obd->obd_name, (s64)elapsed_time / 60, (s64)elapsed_time % 60, atomic_read(&obd->obd_max_recoverable_clients), atomic_read(&obd->obd_connected_clients), obd->obd_stale_clients, obd->obd_stale_clients == 1 ? "was" : "were"); } ldlm_reprocess_recovery_done(obd->obd_namespace); spin_lock(&obd->obd_recovery_task_lock); if (!list_empty(&obd->obd_req_replay_queue) || !list_empty(&obd->obd_lock_replay_queue) || !list_empty(&obd->obd_final_req_queue)) { 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(&obd->obd_recovery_task_lock); LBUG(); } spin_unlock(&obd->obd_recovery_task_lock); obd->obd_recovery_end = ktime_get_seconds(); /* When recovery finished, cleanup orphans on MDS and OST. */ if (obd->obd_type && OBP(obd, postrecov)) { int rc = OBP(obd, postrecov)(obd); if (rc < 0) LCONSOLE_WARN("%s: Post recovery failed, rc %d\n", obd->obd_name, rc); } EXIT; } static void abort_req_replay_queue(struct obd_device *obd) { struct ptlrpc_request *req, *n; LIST_HEAD(abort_list); spin_lock(&obd->obd_recovery_task_lock); list_splice_init(&obd->obd_req_replay_queue, &abort_list); spin_unlock(&obd->obd_recovery_task_lock); 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); target_request_copy_put(req); } } static void abort_lock_replay_queue(struct obd_device *obd) { struct ptlrpc_request *req, *n; LIST_HEAD(abort_list); spin_lock(&obd->obd_recovery_task_lock); list_splice_init(&obd->obd_lock_replay_queue, &abort_list); spin_unlock(&obd->obd_recovery_task_lock); 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)) { DEBUG_REQ(D_ERROR, req, "failed abort_lock_reply; skipping"); } target_request_copy_put(req); } } /* * Called from a cleanup function if the device is being cleaned up * forcefully. The exports should all have been disconnected already, * the only thing left to do is * - clear the recovery flags * - cancel the timer * - free queued requests and replies, but don't send replies * Because the obd_stopping flag is set, no new requests should be received. */ void target_cleanup_recovery(struct obd_device *obd) { struct ptlrpc_request *req, *n; LIST_HEAD(clean_list); spin_lock(&obd->obd_dev_lock); if (!obd->obd_recovering) { spin_unlock(&obd->obd_dev_lock); EXIT; return; } obd->obd_recovering = obd->obd_abort_recovery = 0; spin_unlock(&obd->obd_dev_lock); spin_lock(&obd->obd_recovery_task_lock); target_cancel_recovery_timer(obd); list_splice_init(&obd->obd_req_replay_queue, &clean_list); spin_unlock(&obd->obd_recovery_task_lock); list_for_each_entry_safe(req, n, &clean_list, rq_list) { LASSERT(req->rq_reply_state == NULL); target_exp_dequeue_req_replay(req); target_request_copy_put(req); } spin_lock(&obd->obd_recovery_task_lock); list_splice_init(&obd->obd_lock_replay_queue, &clean_list); list_splice_init(&obd->obd_final_req_queue, &clean_list); spin_unlock(&obd->obd_recovery_task_lock); list_for_each_entry_safe(req, n, &clean_list, rq_list) { LASSERT(req->rq_reply_state == NULL); target_request_copy_put(req); } EXIT; } EXPORT_SYMBOL(target_cleanup_recovery); /* obd_recovery_task_lock should be held */ void target_cancel_recovery_timer(struct obd_device *obd) { CDEBUG(D_HA, "%s: cancel recovery timer\n", obd->obd_name); hrtimer_cancel(&obd->obd_recovery_timer); } static void target_start_recovery_timer(struct obd_device *obd) { ktime_t delay; if (obd->obd_recovery_start != 0) return; spin_lock(&obd->obd_dev_lock); if (!obd->obd_recovering || obd->obd_abort_recovery) { spin_unlock(&obd->obd_dev_lock); return; } LASSERT(obd->obd_recovery_timeout != 0); if (obd->obd_recovery_start != 0) { spin_unlock(&obd->obd_dev_lock); return; } obd->obd_recovery_start = ktime_get_seconds(); delay = ktime_set(obd->obd_recovery_start + obd->obd_recovery_timeout, 0); hrtimer_start(&obd->obd_recovery_timer, delay, HRTIMER_MODE_ABS); spin_unlock(&obd->obd_dev_lock); LCONSOLE_WARN("%s: Will be in recovery for at least %lu:%02lu, or until %d client%s reconnect%s\n", obd->obd_name, obd->obd_recovery_timeout / 60, obd->obd_recovery_timeout % 60, atomic_read(&obd->obd_max_recoverable_clients), (atomic_read(&obd->obd_max_recoverable_clients) == 1) ? "" : "s", (atomic_read(&obd->obd_max_recoverable_clients) == 1) ? "s" : ""); } /** * extend recovery window. * * if @extend is true, extend recovery window to have @dr_timeout remaining * at least; otherwise, make sure the recovery timeout value is not less * than @dr_timeout. */ static void extend_recovery_timer(struct obd_device *obd, time_t dr_timeout, bool extend) { ktime_t left_ns; time_t timeout; time_t left; spin_lock(&obd->obd_dev_lock); if (!obd->obd_recovering || obd->obd_abort_recovery || obd->obd_stopping) { spin_unlock(&obd->obd_dev_lock); return; } LASSERT(obd->obd_recovery_start != 0); left_ns = hrtimer_expires_remaining(&obd->obd_recovery_timer); left = ktime_divns(left_ns, NSEC_PER_SEC); if (extend) { timeout = obd->obd_recovery_timeout; /* dr_timeout will happen after the hrtimer has expired. * Add the excess time to the soft recovery timeout without * exceeding the hard recovery timeout. */ if (dr_timeout > left) { timeout += dr_timeout - left; timeout = min_t(time_t, obd->obd_recovery_time_hard, timeout); } } else { timeout = clamp_t(time_t, dr_timeout, obd->obd_recovery_timeout, obd->obd_recovery_time_hard); } if (timeout == obd->obd_recovery_time_hard) CWARN("%s: extended recovery timer reached hard limit: %ld, extend: %d\n", obd->obd_name, timeout, extend); if (obd->obd_recovery_timeout < timeout) { ktime_t end, now; obd->obd_recovery_timeout = timeout; end = ktime_set(obd->obd_recovery_start + timeout, 0); now = ktime_set(ktime_get_seconds(), 0); left_ns = ktime_sub(end, now); hrtimer_start(&obd->obd_recovery_timer, end, HRTIMER_MODE_ABS); left = ktime_divns(left_ns, NSEC_PER_SEC); } spin_unlock(&obd->obd_dev_lock); CDEBUG(D_HA, "%s: recovery timer will expire in %ld seconds\n", obd->obd_name, left); } /* 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 extended to make sure the client could be reconnected, in the * process, the timeout from the new client should be ignored. */ static void check_and_start_recovery_timer(struct obd_device *obd, struct ptlrpc_request *req, int new_client) { time_t service_time = lustre_msg_get_service_time(req->rq_reqmsg); struct obd_device_target *obt = &obd->u.obt; if (!new_client && service_time) /* * Teach server about old server's estimates, as first guess * at how long new requests will take. */ at_measured(&req->rq_rqbd->rqbd_svcpt->scp_at_estimate, service_time); target_start_recovery_timer(obd); /* * Convert the service time to RPC timeout, * and reuse service_time to limit stack usage. */ service_time = at_est2timeout(service_time); if (OBD_FAIL_CHECK(OBD_FAIL_TGT_SLUGGISH_NET) && service_time < at_extra) service_time = at_extra; /* * We expect other clients to timeout within service_time, then try * to reconnect, then try the failover server. The max delay between * connect attempts is SWITCH_MAX + SWITCH_INC + INITIAL. */ service_time += 2 * INITIAL_CONNECT_TIMEOUT; LASSERT(obt->obt_magic == OBT_MAGIC); service_time += 2 * (CONNECTION_SWITCH_MAX + CONNECTION_SWITCH_INC); if (service_time > obd->obd_recovery_timeout && !new_client) extend_recovery_timer(obd, service_time, false); } /** Health checking routines */ static inline int exp_connect_healthy(struct obd_export *exp) { return exp->exp_in_recovery; } /** if export done req_replay or has replay in queue */ static inline int exp_req_replay_healthy(struct obd_export *exp) { return (!exp->exp_req_replay_needed || atomic_read(&exp->exp_replay_count) > 0); } static inline int exp_req_replay_healthy_or_from_mdt(struct obd_export *exp) { return (exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) || exp_req_replay_healthy(exp); } /** if export done lock_replay or has replay in queue */ static inline int exp_lock_replay_healthy(struct obd_export *exp) { return (!exp->exp_lock_replay_needed || atomic_read(&exp->exp_replay_count) > 0); } static inline int exp_vbr_healthy(struct obd_export *exp) { return !exp->exp_vbr_failed; } static inline int exp_finished(struct obd_export *exp) { return exp->exp_in_recovery && !exp->exp_lock_replay_needed; } static inline int exp_finished_or_from_mdt(struct obd_export *exp) { return (exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) || exp_finished(exp); } 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 req_transno = 0; __u64 update_transno = 0; __u64 next_transno = 0; ENTRY; 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); req_transno = lustre_msg_get_transno(req->rq_reqmsg); } 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; next_transno = obd->obd_next_recovery_transno; CDEBUG(D_HA, "max: %d, connected: %d, completed: %d, queue_len: %d, req_transno: %llu, next_transno: %llu\n", atomic_read(&obd->obd_max_recoverable_clients), connected, completed, queue_len, req_transno, next_transno); if (obd->obd_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 (tdtd != NULL && req != NULL && is_req_replayed_by_update(req)) { LASSERTF(req_transno < next_transno, "req_transno %llu next_transno%llu\n", req_transno, next_transno); CDEBUG(D_HA, "waking for duplicate req (%llu)\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 (%lld)\n", next_transno); wake_up = 1; } else if (queue_len > 0 && queue_len == atomic_read(&obd->obd_req_replay_clients)) { /** handle gaps occured due to lost reply or VBR */ LASSERTF(req_transno >= next_transno, "req_transno: %llu, next_transno: %llu\n", req_transno, next_transno); CDEBUG(D_HA, "%s: waking for gap in transno, VBR is %s (skip: %lld, ql: %d, comp: %d, conn: %d, next: %lld, next_update %lld last_committed: %lld)\n", obd->obd_name, obd->obd_version_recov ? "ON" : "OFF", next_transno, queue_len, completed, connected, 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) { CDEBUG(D_HA, "waking for completed recovery\n"); 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 (%lld)\n", next_transno); obd->obd_next_recovery_transno = req_transno; wake_up = 1; } spin_unlock(&obd->obd_recovery_task_lock); return wake_up; } 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); if (!list_empty(&obd->obd_lock_replay_queue)) { CDEBUG(D_HA, "waking for next lock\n"); wake_up = 1; } 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) { 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; } spin_unlock(&obd->obd_recovery_task_lock); return wake_up; } /** * wait for recovery events, * check its status with help of check_routine * evict dead clients via health_check */ 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; struct target_distribute_txn_data *tdtd; time64_t last = 0; time64_t now; repeat: if (obd->obd_recovering && obd->obd_recovery_start == 0) { now = ktime_get_seconds(); if (now - last > 600) { LCONSOLE_INFO("%s: in recovery but waiting for the first client to connect\n", obd->obd_name); last = now; } } if (obd->obd_recovery_start != 0 && ktime_get_seconds() >= (obd->obd_recovery_start + obd->obd_recovery_time_hard)) { __u64 next_update_transno = 0; /* * Only abort the recovery if there are no update recovery * left in the queue */ spin_lock(&obd->obd_recovery_task_lock); if (lut->lut_tdtd != NULL) { next_update_transno = distribute_txn_get_next_transno(lut->lut_tdtd); tdtd = lut->lut_tdtd; /* * If next_update_transno == 0, it probably because * updatelog retrieve threads did not get any records * yet, let's wait those threads stopped */ if (next_update_transno == 0) { spin_unlock(&obd->obd_recovery_task_lock); wait_event_idle( tdtd->tdtd_recovery_threads_waitq, atomic_read(&tdtd->tdtd_recovery_threads_count) == 0); spin_lock(&obd->obd_recovery_task_lock); next_update_transno = distribute_txn_get_next_transno( lut->lut_tdtd); } } if (next_update_transno != 0 && !obd->obd_abort_recovery) { obd->obd_next_recovery_transno = next_update_transno; spin_unlock(&obd->obd_recovery_task_lock); /* * Disconnect unfinished exports from clients, and * keep connection from MDT to make sure the update * recovery will still keep trying until some one * manually abort the recovery */ class_disconnect_stale_exports(obd, exp_finished_or_from_mdt); /* Abort all of replay & replay lock req from clients */ abort_req_replay_queue(obd); abort_lock_replay_queue(obd); CDEBUG(D_HA, "%s: there are still update replay (%#llx)in the queue.\n", obd->obd_name, next_update_transno); } else { obd->obd_abort_recovery = 1; spin_unlock(&obd->obd_recovery_task_lock); CWARN("%s recovery is aborted by hard timeout\n", obd->obd_name); } } while (wait_event_timeout(obd->obd_next_transno_waitq, check_routine(lut), cfs_time_seconds(60)) == 0) ; /* wait indefinitely for event, but don't trigger watchdog */ if (obd->obd_abort_recovery) { CWARN("recovery is aborted, evict exports in recovery\n"); if (lut->lut_tdtd != NULL) { tdtd = lut->lut_tdtd; /* * Let's wait all of the update log recovery thread * finished */ wait_event_idle( tdtd->tdtd_recovery_threads_waitq, atomic_read(&tdtd->tdtd_recovery_threads_count) == 0); /* Then abort the update recovery list */ dtrq_list_destroy(lut->lut_tdtd); } /** evict exports which didn't finish recovery yet */ class_disconnect_stale_exports(obd, exp_finished); return 1; } else if (obd->obd_recovery_expired && obd->obd_recovery_timeout < obd->obd_recovery_time_hard) { obd->obd_recovery_expired = 0; /** If some clients died being recovered, evict them */ LCONSOLE_WARN("%s: recovery is timed out, evict stale exports\n", obd->obd_name); /** evict cexports with no replay in queue, they are stalled */ class_disconnect_stale_exports(obd, health_check); /** continue with VBR */ spin_lock(&obd->obd_dev_lock); obd->obd_version_recov = 1; spin_unlock(&obd->obd_dev_lock); /** * reset timer, recovery will proceed with versions now, * timeout is set just to handle reconnection delays */ extend_recovery_timer(obd, RECONNECT_DELAY_MAX, true); /** * Wait for recovery events again, after evicting bad clients */ goto repeat; } return 0; } 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(lut, check_for_next_lock, exp_lock_replay_healthy)) abort_lock_replay_queue(obd); spin_lock(&obd->obd_recovery_task_lock); if (!list_empty(&obd->obd_lock_replay_queue)) { req = list_entry(obd->obd_lock_replay_queue.next, struct ptlrpc_request, rq_list); list_del_init(&req->rq_list); spin_unlock(&obd->obd_recovery_task_lock); } else { spin_unlock(&obd->obd_recovery_task_lock); LASSERT(list_empty(&obd->obd_lock_replay_queue)); LASSERT(atomic_read(&obd->obd_lock_replay_clients) == 0); /** evict exports failed VBR */ class_disconnect_stale_exports(obd, exp_vbr_healthy); } return req; } static struct ptlrpc_request *target_next_final_ping(struct obd_device *obd) { struct ptlrpc_request *req = NULL; spin_lock(&obd->obd_recovery_task_lock); if (!list_empty(&obd->obd_final_req_queue)) { req = list_entry(obd->obd_final_req_queue.next, struct ptlrpc_request, rq_list); list_del_init(&req->rq_list); spin_unlock(&obd->obd_recovery_task_lock); 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 { spin_unlock(&obd->obd_recovery_task_lock); } return req; } static void handle_recovery_req(struct ptlrpc_thread *thread, struct ptlrpc_request *req, svc_handler_t handler) { ENTRY; /** * export can be evicted during recovery, no need to handle replays for * it after that, discard such request silently */ if (req->rq_export->exp_disconnected) RETURN_EXIT; req->rq_session.lc_thread = thread; req->rq_svc_thread = thread; req->rq_svc_thread->t_env->le_ses = &req->rq_session; /* thread context */ lu_context_enter(&thread->t_env->le_ctx); (void)handler(req); lu_context_exit(&thread->t_env->le_ctx); req->rq_svc_thread->t_env->le_ses = NULL; /* don't reset timer for final stage */ if (!exp_finished(req->rq_export)) { time_t to = obd_timeout; /** * Add request timeout to the recovery time so next request from * this client may come in recovery time */ if (!AT_OFF) { struct ptlrpc_service_part *svcpt; svcpt = req->rq_rqbd->rqbd_svcpt; /* * If the server sent early reply for this request, * the client will recalculate the timeout according to * current server estimate service time, so we will * use the maxium timeout here for waiting the client * sending the next req */ to = max_t(time_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)), lustre_msg_get_timeout(req->rq_reqmsg)); /* * Add 2 net_latency, one for balance rq_deadline * (see ptl_send_rpc), one for resend the req to server, * Note: client will pack net_latency in replay req * (see ptlrpc_replay_req) */ to += 2 * lustre_msg_get_service_time(req->rq_reqmsg); } extend_recovery_timer(class_exp2obd(req->rq_export), to, true); } 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, atomic_read(&obd->obd_max_recoverable_clients), obd->obd_abort_recovery, obd->obd_recovery_expired); if (!obd->obd_abort_recovery && !obd->obd_recovery_expired) { LASSERT(clnts <= atomic_read(&obd->obd_max_recoverable_clients)); if (clnts + obd->obd_stale_clients < atomic_read(&obd->obd_max_recoverable_clients)) return 0; } if (lut->lut_tdtd != NULL) { if (!lut->lut_tdtd->tdtd_replay_ready && !obd->obd_abort_recovery && !obd->obd_stopping) { /* * 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); CDEBUG(D_HA, "%s update recovery is not ready, extend recovery %lu\n", obd->obd_name, obd->obd_recovery_timeout); return 0; } } 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; } static __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%lld from %s because duplicate update records found", 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", 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 %lld\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_or_from_mdt)) { abort_req_replay_queue(obd); abort_lock_replay_queue(obd); goto abort; } 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, transno); 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 x%llu t%lld from %s", req->rq_xid, 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; int rc; 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); rc = tdtd->tdtd_replay_handler(env, tdtd, dtrq); lu_context_exit(&thread->t_env->le_ctx); extend_recovery_timer(obd, obd_timeout, true); if (rc == 0 && dtrq->dtrq_xid != 0) { CDEBUG(D_HA, "Move x%llu t%llu to finish list\n", dtrq->dtrq_xid, dtrq->dtrq_master_transno); /* Add it to the replay finish list */ spin_lock(&tdtd->tdtd_replay_list_lock); list_add(&dtrq->dtrq_list, &tdtd->tdtd_replay_finish_list); 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 { dtrq_destroy(dtrq); } } else { spin_unlock(&obd->obd_recovery_task_lock); abort: 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; struct obd_device *obd = lut->lut_obd; struct ptlrpc_request *req; struct target_recovery_data *trd = &obd->obd_recovery_data; unsigned long delta; struct lu_env *env; struct ptlrpc_thread *thread = NULL; int rc = 0; ENTRY; unshare_fs_struct(); OBD_ALLOC_PTR(thread); if (thread == NULL) RETURN(-ENOMEM); OBD_ALLOC_PTR(env); if (env == NULL) GOTO(out_thread, rc = -ENOMEM); rc = lu_env_add(env); if (rc) GOTO(out_env, rc); rc = lu_context_init(&env->le_ctx, LCT_MD_THREAD | LCT_DT_THREAD); if (rc) GOTO(out_env_remove, rc); thread->t_env = env; thread->t_id = -1; /* force filter_iobuf_get/put to use local buffers */ env->le_ctx.lc_thread = thread; tgt_io_thread_init(thread); /* init thread_big_cache for IO requests */ CDEBUG(D_HA, "%s: started recovery thread pid %d\n", obd->obd_name, current_pid()); trd->trd_processing_task = current_pid(); spin_lock(&obd->obd_dev_lock); obd->obd_recovering = 1; 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(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 or update */ delta = jiffies; CDEBUG(D_INFO, "1: request replay stage - %d clients from t%llu\n", atomic_read(&obd->obd_req_replay_clients), obd->obd_next_recovery_transno); 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(lut))) { LASSERT(trd->trd_processing_task == current_pid()); DEBUG_REQ(D_HA, req, "processing lock from %s:", libcfs_nid2str(req->rq_peer.nid)); handle_recovery_req(thread, req, trd->trd_recovery_handler); target_request_copy_put(req); obd->obd_replayed_locks++; } /** * The third stage: reply on final pings, at this moment all clients * must have request in final queue */ CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLAY_RECONNECT, cfs_fail_val); CDEBUG(D_INFO, "3: final stage - process recovery completion pings\n"); /** Update server last boot epoch */ tgt_boot_epoch_update(lut); /* * We drop recoverying flag to forward all new requests * to regular mds_handle() since now */ spin_lock(&obd->obd_dev_lock); obd->obd_recovering = obd->obd_abort_recovery = 0; spin_unlock(&obd->obd_dev_lock); spin_lock(&obd->obd_recovery_task_lock); target_cancel_recovery_timer(obd); spin_unlock(&obd->obd_recovery_task_lock); while ((req = target_next_final_ping(obd))) { LASSERT(trd->trd_processing_task == current_pid()); DEBUG_REQ(D_HA, req, "processing final ping from %s:", libcfs_nid2str(req->rq_peer.nid)); handle_recovery_req(thread, req, trd->trd_recovery_handler); /* * Because the waiting client can not send ping to server, * so we need refresh the last_request_time, to avoid the * export is being evicted */ ptlrpc_update_export_timer(req->rq_export, 0); target_request_copy_put(req); } delta = jiffies_to_msecs(jiffies - delta) / MSEC_PER_SEC; CDEBUG(D_INFO, "4: recovery completed in %lus - %d/%d reqs/locks\n", delta, obd->obd_replayed_requests, obd->obd_replayed_locks); if (delta > OBD_RECOVERY_TIME_SOFT) { CWARN("too long recovery - read logs\n"); libcfs_debug_dumplog(); } target_finish_recovery(lut); lu_context_fini(&env->le_ctx); trd->trd_processing_task = 0; complete_all(&trd->trd_finishing); tgt_io_thread_done(thread); out_env_remove: lu_env_remove(env); out_env: OBD_FREE_PTR(env); out_thread: OBD_FREE_PTR(thread); RETURN(rc); } 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; int index; memset(trd, 0, sizeof(*trd)); init_completion(&trd->trd_starting); init_completion(&trd->trd_finishing); trd->trd_recovery_handler = handler; rc = server_name2index(obd->obd_name, &index, NULL); if (rc < 0) return rc; if (!IS_ERR(kthread_run(target_recovery_thread, lut, "tgt_recover_%d", index))) { wait_for_completion(&trd->trd_starting); LASSERT(obd->obd_recovering != 0); } else { rc = -ECHILD; } return rc; } void target_stop_recovery_thread(struct obd_device *obd) { if (obd->obd_recovery_data.trd_processing_task > 0) { struct target_recovery_data *trd = &obd->obd_recovery_data; /** recovery can be done but postrecovery is not yet */ spin_lock(&obd->obd_dev_lock); if (obd->obd_recovering) { CERROR("%s: Aborting recovery\n", obd->obd_name); obd->obd_abort_recovery = 1; wake_up(&obd->obd_next_transno_waitq); } spin_unlock(&obd->obd_dev_lock); wait_for_completion(&trd->trd_finishing); } } EXPORT_SYMBOL(target_stop_recovery_thread); void target_recovery_fini(struct obd_device *obd) { class_disconnect_exports(obd); target_stop_recovery_thread(obd); target_cleanup_recovery(obd); } EXPORT_SYMBOL(target_recovery_fini); static enum hrtimer_restart target_recovery_expired(struct hrtimer *timer) { struct obd_device *obd = container_of(timer, struct obd_device, obd_recovery_timer); CDEBUG(D_HA, "%s: recovery timed out; %d clients are still in recovery after %llu seconds (%d clients connected)\n", obd->obd_name, atomic_read(&obd->obd_lock_replay_clients), ktime_get_real_seconds() - obd->obd_recovery_start, atomic_read(&obd->obd_connected_clients)); obd->obd_recovery_expired = 1; wake_up(&obd->obd_next_transno_waitq); return HRTIMER_NORESTART; } void target_recovery_init(struct lu_target *lut, svc_handler_t handler) { struct obd_device *obd = lut->lut_obd; if (lut->lut_bottom->dd_rdonly) return; if (atomic_read(&obd->obd_max_recoverable_clients) == 0) { /** Update server last boot epoch */ tgt_boot_epoch_update(lut); return; } CDEBUG(D_HA, "RECOVERY: service %s, %d recoverable clients, " "last_transno %llu\n", obd->obd_name, atomic_read(&obd->obd_max_recoverable_clients), obd->obd_last_committed); LASSERT(obd->obd_stopping == 0); obd->obd_next_recovery_transno = obd->obd_last_committed + 1; obd->obd_recovery_start = 0; obd->obd_recovery_end = 0; hrtimer_init(&obd->obd_recovery_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); obd->obd_recovery_timer.function = &target_recovery_expired; target_start_recovery_thread(lut, handler); } EXPORT_SYMBOL(target_recovery_init); static int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req) { struct obd_export *exp = req->rq_export; LASSERT(exp != NULL); if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) { /* client declares he's ready to replay locks */ spin_lock(&exp->exp_lock); if (exp->exp_req_replay_needed) { exp->exp_req_replay_needed = 0; spin_unlock(&exp->exp_lock); LASSERT_ATOMIC_POS(&obd->obd_req_replay_clients); atomic_dec(&obd->obd_req_replay_clients); } else { spin_unlock(&exp->exp_lock); } } if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) { /* * client declares he's ready to complete recovery * so, we put the request on th final queue */ spin_lock(&exp->exp_lock); if (exp->exp_lock_replay_needed) { exp->exp_lock_replay_needed = 0; spin_unlock(&exp->exp_lock); LASSERT_ATOMIC_POS(&obd->obd_lock_replay_clients); atomic_dec(&obd->obd_lock_replay_clients); } else { spin_unlock(&exp->exp_lock); } } return 0; } int target_queue_recovery_request(struct ptlrpc_request *req, struct obd_device *obd) { __u64 transno = lustre_msg_get_transno(req->rq_reqmsg); struct ptlrpc_request *reqiter; int inserted = 0; ENTRY; if (obd->obd_recovery_data.trd_processing_task == current_pid()) { /* Processing the queue right now, don't re-add. */ RETURN(1); } target_process_req_flags(obd, req); if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LOCK_REPLAY_DONE) { if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_TGT_RECOVERY_REQ_RACE))) { if (cfs_fail_val == 1) { cfs_race_state = 1; cfs_fail_val = 0; wake_up(&cfs_race_waitq); schedule_timeout_interruptible( cfs_time_seconds(1)); } } /* * client declares he's ready to complete recovery * so, we put the request on th final queue */ target_request_copy_get(req); DEBUG_REQ(D_HA, req, "queue final req"); wake_up(&obd->obd_next_transno_waitq); spin_lock(&obd->obd_recovery_task_lock); if (obd->obd_recovering) { struct ptlrpc_request *tmp; struct ptlrpc_request *duplicate = NULL; if (likely(!req->rq_export->exp_replay_done)) { req->rq_export->exp_replay_done = 1; list_add_tail(&req->rq_list, &obd->obd_final_req_queue); spin_unlock(&obd->obd_recovery_task_lock); RETURN(0); } /* * XXX O(n), but only happens if final ping is * timed out, probably reorganize the list as * a hash list later */ list_for_each_entry_safe(reqiter, tmp, &obd->obd_final_req_queue, rq_list) { if (reqiter->rq_export == req->rq_export) { list_del_init(&reqiter->rq_list); duplicate = reqiter; break; } } list_add_tail(&req->rq_list, &obd->obd_final_req_queue); req->rq_export->exp_replay_done = 1; spin_unlock(&obd->obd_recovery_task_lock); if (duplicate != NULL) { DEBUG_REQ(D_HA, duplicate, "put prev final req"); target_request_copy_put(duplicate); } RETURN(0); } else { spin_unlock(&obd->obd_recovery_task_lock); target_request_copy_put(req); RETURN(obd->obd_stopping ? -ENOTCONN : 1); } } if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REQ_REPLAY_DONE) { /* client declares he's ready to replay locks */ target_request_copy_get(req); DEBUG_REQ(D_HA, req, "queue lock replay req"); wake_up(&obd->obd_next_transno_waitq); spin_lock(&obd->obd_recovery_task_lock); LASSERT(obd->obd_recovering); /* usually due to recovery abort */ if (!req->rq_export->exp_in_recovery) { spin_unlock(&obd->obd_recovery_task_lock); target_request_copy_put(req); RETURN(-ENOTCONN); } LASSERT(req->rq_export->exp_lock_replay_needed); list_add_tail(&req->rq_list, &obd->obd_lock_replay_queue); spin_unlock(&obd->obd_recovery_task_lock); RETURN(0); } /* * CAVEAT EMPTOR: The incoming request message has been swabbed * (i.e. buflens etc are in my own byte order), but type-dependent * buffers (eg mdt_body, ost_body etc) have NOT been swabbed. */ if (!transno) { INIT_LIST_HEAD(&req->rq_list); DEBUG_REQ(D_HA, req, "not queueing"); RETURN(1); } /* * If we're processing the queue, we want don't want to queue this * message. * * Also, if this request has a transno less than the one we're waiting * for, we should process it now. It could (and currently always will) * be an open request for a descriptor that was opened some time ago. * * Also, a resent, replayed request that has already been * handled will pass through here and be processed immediately. */ CDEBUG(D_HA, "Next recovery transno: %llu, current: %llu, 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 && !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); RETURN(1); } spin_unlock(&obd->obd_recovery_task_lock); if (OBD_FAIL_CHECK(OBD_FAIL_TGT_REPLAY_DROP)) RETURN(0); target_request_copy_get(req); if (!req->rq_export->exp_in_recovery) { target_request_copy_put(req); RETURN(-ENOTCONN); } LASSERT(req->rq_export->exp_req_replay_needed); if (target_exp_enqueue_req_replay(req)) { DEBUG_REQ(D_ERROR, req, "dropping resent queued req"); target_request_copy_put(req); RETURN(0); } /* XXX O(n^2) */ spin_lock(&obd->obd_recovery_task_lock); LASSERT(obd->obd_recovering); list_for_each_entry(reqiter, &obd->obd_req_replay_queue, rq_list) { if (lustre_msg_get_transno(reqiter->rq_reqmsg) > transno) { list_add_tail(&req->rq_list, &reqiter->rq_list); inserted = 1; goto added; } 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(&obd->obd_recovery_task_lock); target_exp_dequeue_req_replay(req); target_request_copy_put(req); RETURN(0); } } added: if (!inserted) list_add_tail(&req->rq_list, &obd->obd_req_replay_queue); obd->obd_requests_queued_for_recovery++; spin_unlock(&obd->obd_recovery_task_lock); wake_up(&obd->obd_next_transno_waitq); RETURN(0); } void target_committed_to_req(struct ptlrpc_request *req) { struct obd_export *exp = req->rq_export; if (!exp->exp_obd->obd_no_transno && req->rq_repmsg != NULL) lustre_msg_set_last_committed(req->rq_repmsg, exp->exp_last_committed); else 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 %llu, transno %llu, xid %llu\n", exp->exp_last_committed, req->rq_transno, req->rq_xid); } #endif /* HAVE_SERVER_SUPPORT */ /** * Packs current SLV and Limit into \a req. */ int target_pack_pool_reply(struct ptlrpc_request *req) { struct obd_device *obd; ENTRY; /* * Check that we still have all structures alive as this may * be some late RPC at 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); } /* OBD is alive here as export is alive, which we checked above. */ obd = req->rq_export->exp_obd; 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); } 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; } /* * We can have a null rq_reqmsg in the event of bad signature or * no context when unwrapping */ if (req->rq_reqmsg && 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); } DEBUG_REQ(D_NET, req, "sending reply"); return ptlrpc_send_reply(req, PTLRPC_REPLY_MAYBE_DIFFICULT); } void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) { struct ptlrpc_service_part *svcpt; int netrc; struct ptlrpc_reply_state *rs; struct obd_export *exp; ENTRY; if (req->rq_no_reply) { EXIT; return; } svcpt = req->rq_rqbd->rqbd_svcpt; rs = req->rq_reply_state; if (rs == NULL || !rs->rs_difficult) { /* no notifiers */ target_send_reply_msg(req, rc, fail_id); EXIT; return; } /* must be an export if locks saved */ LASSERT(req->rq_export != NULL); /* req/reply consistent */ LASSERT(rs->rs_svcpt == svcpt); /* "fresh" reply */ LASSERT(!rs->rs_scheduled); LASSERT(!rs->rs_scheduled_ever); LASSERT(!rs->rs_handled); LASSERT(!rs->rs_on_net); LASSERT(rs->rs_export == NULL); LASSERT(list_empty(&rs->rs_obd_list)); LASSERT(list_empty(&rs->rs_exp_list)); exp = class_export_get(req->rq_export); /* disable reply scheduling while I'm setting up */ rs->rs_scheduled = 1; rs->rs_on_net = 1; rs->rs_xid = req->rq_xid; rs->rs_transno = req->rq_transno; rs->rs_export = exp; rs->rs_opc = lustre_msg_get_opc(req->rq_reqmsg); spin_lock(&exp->exp_uncommitted_replies_lock); CDEBUG(D_NET, "rs transno = %llu, last committed = %llu\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, &exp->exp_uncommitted_replies); } spin_unlock(&exp->exp_uncommitted_replies_lock); 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(&svcpt->scp_rep_lock); atomic_inc(&svcpt->scp_nreps_difficult); if (netrc != 0) { /* * error sending: reply is off the net. Also we need +1 * reply ref until ptlrpc_handle_rs() is done * with the reply state (if the send was successful, there * would have been +1 ref for the net, which * reply_out_callback leaves alone) */ rs->rs_on_net = 0; ptlrpc_rs_addref(rs); } 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, &svcpt->scp_rep_active); rs->rs_scheduled = 0; /* allow notifier to schedule */ } spin_unlock(&rs->rs_lock); spin_unlock(&svcpt->scp_rep_lock); EXIT; } enum ldlm_mode lck_compat_array[] = { [LCK_EX] = LCK_COMPAT_EX, [LCK_PW] = LCK_COMPAT_PW, [LCK_PR] = LCK_COMPAT_PR, [LCK_CW] = LCK_COMPAT_CW, [LCK_CR] = LCK_COMPAT_CR, [LCK_NL] = LCK_COMPAT_NL, [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(enum ldlm_error error) { int result; switch (error) { case ELDLM_OK: case ELDLM_LOCK_MATCHED: 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 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 enum ldlm_error. */ enum ldlm_error 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; } #if LUSTRE_TRACKS_LOCK_EXP_REFS void ldlm_dump_export_locks(struct obd_export *exp) { spin_lock(&exp->exp_locks_list_guard); if (!list_empty(&exp->exp_locks_list)) { struct ldlm_lock *lock; CERROR("dumping locks for export %p, ignore if the unmount doesn't hang\n", exp); list_for_each_entry(lock, &exp->exp_locks_list, l_exp_refs_link) LDLM_ERROR(lock, "lock:"); } spin_unlock(&exp->exp_locks_list_guard); } #endif #ifdef HAVE_SERVER_SUPPORT static inline const char *bulk2type(struct ptlrpc_request *req) { if (req->rq_bulk_read) return "READ"; if (req->rq_bulk_write) return "WRITE"; return "UNKNOWN"; } int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc) { struct ptlrpc_request *req = desc->bd_req; time64_t start = ktime_get_seconds(); time64_t deadline; int rc = 0; ENTRY; /* If there is eviction in progress, wait for it to finish. */ wait_event_idle( exp->exp_obd->obd_evict_inprogress_waitq, !atomic_read(&exp->exp_obd->obd_evict_inprogress)); /* Check if client was evicted or reconnected already. */ if (exp->exp_failed || exp->exp_conn_cnt > lustre_msg_get_conn_cnt(req->rq_reqmsg)) { rc = -ENOTCONN; } else { if (req->rq_bulk_read) rc = sptlrpc_svc_wrap_bulk(req, desc); if (OCD_HAS_FLAG(&exp->exp_connect_data, BULK_MBITS)) 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); } if (rc < 0) { DEBUG_REQ(D_ERROR, req, "bulk %s failed: rc = %d", bulk2type(req), rc); RETURN(rc); } if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) { ptlrpc_abort_bulk(desc); RETURN(0); } /* limit actual bulk transfer to bulk_timeout seconds */ deadline = start + bulk_timeout; if (deadline > req->rq_deadline) deadline = req->rq_deadline; do { time64_t timeoutl = deadline - ktime_get_seconds(); time64_t rq_deadline; while (timeoutl >= 0 && wait_event_idle_timeout( desc->bd_waitq, !ptlrpc_server_bulk_active(desc) || exp->exp_failed || exp->exp_conn_cnt > lustre_msg_get_conn_cnt(req->rq_reqmsg), timeoutl ? cfs_time_seconds(1) : 1) == 0) timeoutl -= 1; rc = timeoutl < 0 ? -ETIMEDOUT : 0; /* Wait again if we changed rq_deadline. */ rq_deadline = READ_ONCE(req->rq_deadline); deadline = start + bulk_timeout; if (deadline > rq_deadline) deadline = rq_deadline; } while (rc == -ETIMEDOUT && deadline > ktime_get_seconds()); if (rc == -ETIMEDOUT) { DEBUG_REQ(D_ERROR, req, "timeout on bulk %s after %lld%+llds", bulk2type(req), deadline - start, ktime_get_real_seconds() - deadline); ptlrpc_abort_bulk(desc); } else if (exp->exp_failed) { DEBUG_REQ(D_ERROR, req, "Eviction on bulk %s", bulk2type(req)); rc = -ENOTCONN; ptlrpc_abort_bulk(desc); } else if (exp->exp_conn_cnt > lustre_msg_get_conn_cnt(req->rq_reqmsg)) { DEBUG_REQ(D_ERROR, req, "Reconnect on bulk %s", bulk2type(req)); /* We don't reply anyway. */ rc = -ETIMEDOUT; ptlrpc_abort_bulk(desc); } else if (desc->bd_failure) { DEBUG_REQ(D_ERROR, req, "network error on bulk %s", bulk2type(req)); /* XXX should this be a different errno? */ rc = -ETIMEDOUT; } else { if (req->rq_bulk_write) rc = sptlrpc_svc_unwrap_bulk(req, desc); if (rc == 0 && desc->bd_nob_transferred != desc->bd_nob) { DEBUG_REQ(D_ERROR, req, "truncated bulk %s %d(%d)", bulk2type(req), desc->bd_nob_transferred, desc->bd_nob); /* XXX should this be a different errno? */ rc = -ETIMEDOUT; } } RETURN(rc); } EXPORT_SYMBOL(target_bulk_io); #endif /* HAVE_SERVER_SUPPORT */