X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lib.c;h=d747a183ab42cd143258668bf38948a143bbed85;hb=8f1c8dbe2389e1ef1e1d3387e343fb9a1bb84198;hp=c22eccb4ee79b09c2be53d4c530a4573899e41f6;hpb=4a540ca0c37a5302c9696b2b626cbb61384c60be;p=fs%2Flustre-release.git diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c index c22eccb..d747a18 100644 --- a/lustre/ldlm/ldlm_lib.c +++ b/lustre/ldlm/ldlm_lib.c @@ -27,7 +27,7 @@ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2010, 2013, Intel Corporation. + * Copyright (c) 2010, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -43,11 +43,8 @@ #define DEBUG_SUBSYSTEM S_LDLM -#ifdef __KERNEL__ -# include -#else -# include -#endif +#include +#include #include #include #include @@ -85,11 +82,11 @@ static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid, } spin_lock(&imp->imp_lock); - cfs_list_for_each_entry(item, &imp->imp_conn_list, oic_item) { + list_for_each_entry(item, &imp->imp_conn_list, oic_item) { if (obd_uuid_equals(uuid, &item->oic_uuid)) { if (priority) { - cfs_list_del(&item->oic_item); - cfs_list_add(&item->oic_item, + list_del(&item->oic_item); + list_add(&item->oic_item, &imp->imp_conn_list); item->oic_last_attempt = 0; } @@ -106,9 +103,9 @@ static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid, imp_conn->oic_uuid = *uuid; imp_conn->oic_last_attempt = 0; if (priority) - cfs_list_add(&imp_conn->oic_item, &imp->imp_conn_list); + list_add(&imp_conn->oic_item, &imp->imp_conn_list); else - cfs_list_add_tail(&imp_conn->oic_item, + 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, @@ -148,12 +145,12 @@ int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid) ENTRY; spin_lock(&imp->imp_lock); - if (cfs_list_empty(&imp->imp_conn_list)) { + if (list_empty(&imp->imp_conn_list)) { LASSERT(!imp->imp_connection); GOTO(out, rc); } - cfs_list_for_each_entry(imp_conn, &imp->imp_conn_list, oic_item) { + 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); @@ -179,7 +176,7 @@ int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid) } } - cfs_list_del(&imp_conn->oic_item); + 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", @@ -207,7 +204,7 @@ int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer, ENTRY; spin_lock(&imp->imp_lock); - cfs_list_for_each_entry(conn, &imp->imp_conn_list, oic_item) { + 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; @@ -346,18 +343,18 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) min_t(unsigned int, LUSTRE_CFG_BUFLEN(lcfg, 2), sizeof(server_uuid))); - cli->cl_dirty = 0; - cli->cl_avail_grant = 0; - /* FIXME: Should limit this for the sum of all cl_dirty_max. */ - cli->cl_dirty_max = OSC_MAX_DIRTY_DEFAULT * 1024 * 1024; - if (cli->cl_dirty_max >> PAGE_CACHE_SHIFT > totalram_pages / 8) - cli->cl_dirty_max = totalram_pages << (PAGE_CACHE_SHIFT - 3); - CFS_INIT_LIST_HEAD(&cli->cl_cache_waiters); - CFS_INIT_LIST_HEAD(&cli->cl_loi_ready_list); - CFS_INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list); - CFS_INIT_LIST_HEAD(&cli->cl_loi_write_list); - CFS_INIT_LIST_HEAD(&cli->cl_loi_read_list); - client_obd_list_lock_init(&cli->cl_loi_list_lock); + cli->cl_dirty_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; @@ -371,13 +368,13 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) spin_lock_init(&cli->cl_write_offset_hist.oh_lock); /* lru for osc. */ - CFS_INIT_LIST_HEAD(&cli->cl_lru_osc); + INIT_LIST_HEAD(&cli->cl_lru_osc); atomic_set(&cli->cl_lru_shrinkers, 0); - atomic_set(&cli->cl_lru_busy, 0); - atomic_set(&cli->cl_lru_in_list, 0); - CFS_INIT_LIST_HEAD(&cli->cl_lru_list); - client_obd_list_lock_init(&cli->cl_lru_list_lock); - atomic_set(&cli->cl_unstable_count, 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_waitqueue_head(&cli->cl_destroy_waitq); atomic_set(&cli->cl_destroy_in_flight, 0); @@ -512,7 +509,7 @@ int client_connect_import(const struct lu_env *env, *exp = NULL; down_write(&cli->cl_sem); - if (cli->cl_conn_count > 0 ) + if (cli->cl_conn_count > 0) GOTO(out_sem, rc = -EALREADY); rc = class_connect(&conn, obd, cluuid); @@ -584,17 +581,17 @@ int client_disconnect_export(struct obd_export *exp) imp = cli->cl_import; down_write(&cli->cl_sem); - CDEBUG(D_INFO, "disconnect %s - %d\n", obd->obd_name, - cli->cl_conn_count); + CDEBUG(D_INFO, "disconnect %s - %zu\n", obd->obd_name, + cli->cl_conn_count); - if (!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) + if (cli->cl_conn_count != 0) GOTO(out_disconnect, rc = 0); /* Mark import deactivated now, so we don't try to reconnect if any @@ -656,15 +653,15 @@ int server_disconnect_export(struct obd_export *exp) /* complete all outstanding replies */ spin_lock(&exp->exp_lock); - while (!cfs_list_empty(&exp->exp_outstanding_replies)) { + while (!list_empty(&exp->exp_outstanding_replies)) { struct ptlrpc_reply_state *rs = - cfs_list_entry(exp->exp_outstanding_replies.next, + 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); - cfs_list_del_init(&rs->rs_exp_list); + list_del_init(&rs->rs_exp_list); spin_lock(&rs->rs_lock); ptlrpc_schedule_difficult_reply(rs); spin_unlock(&rs->rs_lock); @@ -756,19 +753,10 @@ void target_client_add_cb(struct obd_device *obd, __u64 transno, void *cb_data, spin_unlock(&exp->exp_lock); class_export_cb_put(exp); } -EXPORT_SYMBOL(target_client_add_cb); -#ifdef __KERNEL__ static void check_and_start_recovery_timer(struct obd_device *obd, struct ptlrpc_request *req, int new_client); -#else -static inline void -check_and_start_recovery_timer(struct obd_device *obd, - struct ptlrpc_request *req, int new_client) -{ -} -#endif int target_handle_connect(struct ptlrpc_request *req) { @@ -883,6 +871,20 @@ 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 || @@ -912,6 +914,38 @@ int target_handle_connect(struct ptlrpc_request *req) mds_conn = (data->ocd_connect_flags & OBD_CONNECT_MDS) != 0; lw_client = (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) != 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. */ + 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. */ @@ -1045,49 +1079,35 @@ 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); - } - if (rc) - GOTO(out, rc); - -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0) - /* 2.2.0 clients always swab nidtbl entries due to a bug, so server - * will do the swabbing for if the client is using the same endianness. - * - * This fixup is version-limited, because we don't want to carry the - * OBD_CONNECT_MNE_SWAB flag around forever, just so long as we need - * interop with unpatched 2.2 clients. For newer clients, servers - * will never do MNE swabbing, let the client handle that. LU-1644 */ - export->exp_need_mne_swab = !ptlrpc_req_need_swab(req) && - !(data->ocd_connect_flags & OBD_CONNECT_MNE_SWAB); -#else -#warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and exp_need_mne_swab" -#endif + 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); + } + if (rc) + GOTO(out, rc); - LASSERT(target->u.obt.obt_magic == OBT_MAGIC); - data->ocd_instance = target->u.obt.obt_instance; + 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. */ @@ -1136,7 +1156,7 @@ dont_check_exports: spin_unlock(&export->exp_lock); spin_lock(&target->obd_dev_lock); - cfs_list_del_init(&export->exp_obd_chain_timed); + list_del_init(&export->exp_obd_chain_timed); spin_unlock(&target->obd_dev_lock); } else { spin_unlock(&export->exp_lock); @@ -1145,7 +1165,7 @@ dont_check_exports: if (export->exp_connection != NULL) { /* Check to see if connection came from another NID. */ if ((export->exp_connection->c_peer.nid != req->rq_peer.nid) && - !cfs_hlist_unhashed(&export->exp_nid_hash)) + !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); @@ -1156,7 +1176,7 @@ dont_check_exports: export->exp_connection = ptlrpc_connection_get(req->rq_peer, req->rq_self, &remote_uuid); - if (cfs_hlist_unhashed(&export->exp_nid_hash)) { + 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); @@ -1231,17 +1251,12 @@ dont_check_exports: * 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)) + if (data->ocd_connect_flags & OBD_CONNECT_AT) 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; + revimp->imp_msghdr_flags |= MSGHDR_CKSUM_INCOMPAT18; rc = sptlrpc_import_sec_adapt(revimp, req->rq_svc_ctx, &req->rq_flvr); if (rc) { @@ -1275,7 +1290,6 @@ out: req->rq_status = rc; RETURN(rc); } -EXPORT_SYMBOL(target_handle_connect); int target_handle_disconnect(struct ptlrpc_request *req) { @@ -1291,7 +1305,6 @@ int target_handle_disconnect(struct ptlrpc_request *req) RETURN(0); } -EXPORT_SYMBOL(target_handle_disconnect); void target_destroy_export(struct obd_export *exp) { @@ -1320,8 +1333,8 @@ EXPORT_SYMBOL(target_destroy_export); static void target_request_copy_get(struct ptlrpc_request *req) { class_export_rpc_inc(req->rq_export); - LASSERT(cfs_list_empty(&req->rq_list)); - CFS_INIT_LIST_HEAD(&req->rq_replay_list); + 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); @@ -1331,7 +1344,7 @@ static void target_request_copy_get(struct ptlrpc_request *req) static void target_request_copy_put(struct ptlrpc_request *req) { - LASSERT(cfs_list_empty(&req->rq_replay_list)); + LASSERT(list_empty(&req->rq_replay_list)); LASSERT_ATOMIC_POS(&req->rq_export->exp_replay_count); atomic_dec(&req->rq_export->exp_replay_count); @@ -1349,7 +1362,7 @@ static int target_exp_enqueue_req_replay(struct ptlrpc_request *req) LASSERT(exp); spin_lock(&exp->exp_lock); - cfs_list_for_each_entry(reqiter, &exp->exp_req_replay_queue, + list_for_each_entry(reqiter, &exp->exp_req_replay_queue, rq_replay_list) { if (lustre_msg_get_transno(reqiter->rq_reqmsg) == transno) { dup = 1; @@ -1364,7 +1377,7 @@ static int target_exp_enqueue_req_replay(struct ptlrpc_request *req) CERROR("invalid flags %x of resent replay\n", lustre_msg_get_flags(req->rq_reqmsg)); } else { - cfs_list_add_tail(&req->rq_replay_list, + list_add_tail(&req->rq_replay_list, &exp->exp_req_replay_queue); } @@ -1374,15 +1387,14 @@ static int target_exp_enqueue_req_replay(struct ptlrpc_request *req) static void target_exp_dequeue_req_replay(struct ptlrpc_request *req) { - LASSERT(!cfs_list_empty(&req->rq_replay_list)); + LASSERT(!list_empty(&req->rq_replay_list)); LASSERT(req->rq_export); spin_lock(&req->rq_export->exp_lock); - cfs_list_del_init(&req->rq_replay_list); + list_del_init(&req->rq_replay_list); spin_unlock(&req->rq_export->exp_lock); } -#ifdef __KERNEL__ static void target_finish_recovery(struct obd_device *obd) { ENTRY; @@ -1402,15 +1414,15 @@ static void target_finish_recovery(struct obd_device *obd) ldlm_reprocess_all_ns(obd->obd_namespace); spin_lock(&obd->obd_recovery_task_lock); - if (!cfs_list_empty(&obd->obd_req_replay_queue) || - !cfs_list_empty(&obd->obd_lock_replay_queue) || - !cfs_list_empty(&obd->obd_final_req_queue)) { + 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, - cfs_list_empty(&obd->obd_req_replay_queue) ? "" : "req ", - cfs_list_empty(&obd->obd_lock_replay_queue) ? \ + list_empty(&obd->obd_req_replay_queue) ? "" : "req ", + list_empty(&obd->obd_lock_replay_queue) ? \ "" : "lock ", - cfs_list_empty(&obd->obd_final_req_queue) ? \ + list_empty(&obd->obd_final_req_queue) ? \ "" : "final "); spin_unlock(&obd->obd_recovery_task_lock); LBUG(); @@ -1432,13 +1444,13 @@ static void target_finish_recovery(struct obd_device *obd) static void abort_req_replay_queue(struct obd_device *obd) { struct ptlrpc_request *req, *n; - cfs_list_t abort_list; + struct list_head abort_list; - CFS_INIT_LIST_HEAD(&abort_list); + INIT_LIST_HEAD(&abort_list); spin_lock(&obd->obd_recovery_task_lock); - cfs_list_splice_init(&obd->obd_req_replay_queue, &abort_list); + list_splice_init(&obd->obd_req_replay_queue, &abort_list); spin_unlock(&obd->obd_recovery_task_lock); - cfs_list_for_each_entry_safe(req, n, &abort_list, rq_list) { + 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)) { @@ -1453,13 +1465,13 @@ static void abort_req_replay_queue(struct obd_device *obd) static void abort_lock_replay_queue(struct obd_device *obd) { struct ptlrpc_request *req, *n; - cfs_list_t abort_list; + struct list_head abort_list; - CFS_INIT_LIST_HEAD(&abort_list); + INIT_LIST_HEAD(&abort_list); spin_lock(&obd->obd_recovery_task_lock); - cfs_list_splice_init(&obd->obd_lock_replay_queue, &abort_list); + list_splice_init(&obd->obd_lock_replay_queue, &abort_list); spin_unlock(&obd->obd_recovery_task_lock); - cfs_list_for_each_entry_safe(req, n, &abort_list, rq_list){ + 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)) { @@ -1482,10 +1494,10 @@ static void abort_lock_replay_queue(struct obd_device *obd) void target_cleanup_recovery(struct obd_device *obd) { struct ptlrpc_request *req, *n; - cfs_list_t clean_list; + struct list_head clean_list; ENTRY; - CFS_INIT_LIST_HEAD(&clean_list); + INIT_LIST_HEAD(&clean_list); spin_lock(&obd->obd_dev_lock); if (!obd->obd_recovering) { spin_unlock(&obd->obd_dev_lock); @@ -1497,22 +1509,22 @@ void target_cleanup_recovery(struct obd_device *obd) spin_lock(&obd->obd_recovery_task_lock); target_cancel_recovery_timer(obd); - cfs_list_splice_init(&obd->obd_req_replay_queue, &clean_list); + list_splice_init(&obd->obd_req_replay_queue, &clean_list); spin_unlock(&obd->obd_recovery_task_lock); - cfs_list_for_each_entry_safe(req, n, &clean_list, rq_list) { - LASSERT(req->rq_reply_state == 0); + 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); - cfs_list_splice_init(&obd->obd_lock_replay_queue, &clean_list); - cfs_list_splice_init(&obd->obd_final_req_queue, &clean_list); + 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); - cfs_list_for_each_entry_safe(req, n, &clean_list, rq_list){ - LASSERT(req->rq_reply_state == 0); + list_for_each_entry_safe(req, n, &clean_list, rq_list) { + LASSERT(req->rq_reply_state == NULL); target_request_copy_put(req); } @@ -1526,7 +1538,6 @@ void target_cancel_recovery_timer(struct obd_device *obd) CDEBUG(D_HA, "%s: cancel recovery timer\n", obd->obd_name); cfs_timer_disarm(&obd->obd_recovery_timer); } -EXPORT_SYMBOL(target_cancel_recovery_timer); static void target_start_recovery_timer(struct obd_device *obd) { @@ -1590,22 +1601,20 @@ static void extend_recovery_timer(struct obd_device *obd, int drt, bool extend) to += drt - left; } else if (!extend && (drt > to)) { to = drt; - /* reduce drt by already passed time */ - drt -= obd->obd_recovery_timeout - left; } if (to > obd->obd_recovery_time_hard) to = obd->obd_recovery_time_hard; - if (obd->obd_recovery_timeout < to || - obd->obd_recovery_timeout == obd->obd_recovery_time_hard) { + if (obd->obd_recovery_timeout < to) { obd->obd_recovery_timeout = to; - cfs_timer_arm(&obd->obd_recovery_timer, - cfs_time_shift(drt)); + end = obd->obd_recovery_start + to; + cfs_timer_arm(&obd->obd_recovery_timer, + cfs_time_shift(end - now)); } spin_unlock(&obd->obd_dev_lock); CDEBUG(D_HA, "%s: recovery timer will expire in %u seconds\n", - obd->obd_name, (unsigned)drt); + obd->obd_name, (unsigned)cfs_time_sub(end, now)); } /* Reset the timer with each new client connection */ @@ -1699,8 +1708,8 @@ static int check_for_next_transno(struct obd_device *obd) ENTRY; spin_lock(&obd->obd_recovery_task_lock); - if (!cfs_list_empty(&obd->obd_req_replay_queue)) { - req = cfs_list_entry(obd->obd_req_replay_queue.next, + 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); } else { @@ -1763,7 +1772,7 @@ static int check_for_next_lock(struct obd_device *obd) int wake_up = 0; spin_lock(&obd->obd_recovery_task_lock); - if (!cfs_list_empty(&obd->obd_lock_replay_queue)) { + 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) { @@ -1791,7 +1800,17 @@ static int target_recovery_overseer(struct obd_device *obd, int (*health_check)(struct obd_export *)) { repeat: - wait_event(obd->obd_next_transno_waitq, check_routine(obd)); + if ((obd->obd_recovery_start != 0) && (cfs_time_current_sec() >= + (obd->obd_recovery_start + obd->obd_recovery_time_hard))) { + CWARN("recovery is aborted by hard timeout\n"); + obd->obd_abort_recovery = 1; + } + + while (wait_event_timeout(obd->obd_next_transno_waitq, + check_routine(obd), + msecs_to_jiffies(60 * MSEC_PER_SEC)) == 0) + /* wait indefinitely for event, but don't trigger watchdog */; + if (obd->obd_abort_recovery) { CWARN("recovery is aborted, evict exports in recovery\n"); /** evict exports which didn't finish recovery yet */ @@ -1821,28 +1840,35 @@ repeat: static struct ptlrpc_request *target_next_replay_req(struct obd_device *obd) { - struct ptlrpc_request *req = NULL; - ENTRY; + struct ptlrpc_request *req = NULL; + ENTRY; - CDEBUG(D_HA, "Waiting for transno "LPD64"\n", - obd->obd_next_recovery_transno); + CDEBUG(D_HA, "Waiting for transno "LPD64"\n", + obd->obd_next_recovery_transno); - if (target_recovery_overseer(obd, check_for_next_transno, - exp_req_replay_healthy)) { - abort_req_replay_queue(obd); - abort_lock_replay_queue(obd); - } + 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 (!cfs_list_empty(&obd->obd_req_replay_queue)) { - req = cfs_list_entry(obd->obd_req_replay_queue.next, + if (!list_empty(&obd->obd_req_replay_queue)) { + req = list_entry(obd->obd_req_replay_queue.next, struct ptlrpc_request, rq_list); - cfs_list_del_init(&req->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(cfs_list_empty(&obd->obd_req_replay_queue)); + 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); @@ -1860,14 +1886,14 @@ static struct ptlrpc_request *target_next_replay_lock(struct obd_device *obd) abort_lock_replay_queue(obd); spin_lock(&obd->obd_recovery_task_lock); - if (!cfs_list_empty(&obd->obd_lock_replay_queue)) { - req = cfs_list_entry(obd->obd_lock_replay_queue.next, + if (!list_empty(&obd->obd_lock_replay_queue)) { + req = list_entry(obd->obd_lock_replay_queue.next, struct ptlrpc_request, rq_list); - cfs_list_del_init(&req->rq_list); + list_del_init(&req->rq_list); spin_unlock(&obd->obd_recovery_task_lock); } else { spin_unlock(&obd->obd_recovery_task_lock); - LASSERT(cfs_list_empty(&obd->obd_lock_replay_queue)); + 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); @@ -1880,10 +1906,10 @@ 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 (!cfs_list_empty(&obd->obd_final_req_queue)) { - req = cfs_list_entry(obd->obd_final_req_queue.next, + if (!list_empty(&obd->obd_final_req_queue)) { + req = list_entry(obd->obd_final_req_queue.next, struct ptlrpc_request, rq_list); - cfs_list_del_init(&req->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); @@ -2044,6 +2070,7 @@ static int target_recovery_thread(void *arg) * 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); @@ -2061,10 +2088,14 @@ static int target_recovery_thread(void *arg) libcfs_nid2str(req->rq_peer.nid)); handle_recovery_req(thread, req, trd->trd_recovery_handler); - target_request_copy_put(req); - } + /* 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 - delta) / HZ; + 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) { @@ -2167,7 +2198,6 @@ void target_recovery_init(struct lu_target *lut, svc_handler_t handler) } EXPORT_SYMBOL(target_recovery_init); -#endif /* __KERNEL__ */ static int target_process_req_flags(struct obd_device *obd, struct ptlrpc_request *req) @@ -2207,10 +2237,10 @@ static int target_process_req_flags(struct obd_device *obd, int target_queue_recovery_request(struct ptlrpc_request *req, struct obd_device *obd) { - cfs_list_t *tmp; - int inserted = 0; __u64 transno = lustre_msg_get_transno(req->rq_reqmsg); - ENTRY; + 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. */ @@ -2227,7 +2257,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req, wake_up(&obd->obd_next_transno_waitq); spin_lock(&obd->obd_recovery_task_lock); if (obd->obd_recovering) { - cfs_list_add_tail(&req->rq_list, + list_add_tail(&req->rq_list, &obd->obd_final_req_queue); } else { spin_unlock(&obd->obd_recovery_task_lock); @@ -2251,7 +2281,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req, RETURN(-ENOTCONN); } LASSERT(req->rq_export->exp_lock_replay_needed); - cfs_list_add_tail(&req->rq_list, &obd->obd_lock_replay_queue); + list_add_tail(&req->rq_list, &obd->obd_lock_replay_queue); spin_unlock(&obd->obd_recovery_task_lock); RETURN(0); } @@ -2261,7 +2291,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req, * buffers (eg mdt_body, ost_body etc) have NOT been swabbed. */ if (!transno) { - CFS_INIT_LIST_HEAD(&req->rq_list); + INIT_LIST_HEAD(&req->rq_list); DEBUG_REQ(D_HA, req, "not queueing"); RETURN(1); } @@ -2282,7 +2312,7 @@ int target_queue_recovery_request(struct ptlrpc_request *req, spin_lock(&obd->obd_recovery_task_lock); if (transno < obd->obd_next_recovery_transno) { /* Processing the queue right now, don't re-add. */ - LASSERT(cfs_list_empty(&req->rq_list)); + LASSERT(list_empty(&req->rq_list)); spin_unlock(&obd->obd_recovery_task_lock); RETURN(1); } @@ -2304,18 +2334,15 @@ int target_queue_recovery_request(struct ptlrpc_request *req, RETURN(0); } - /* XXX O(n^2) */ + /* XXX O(n^2) */ spin_lock(&obd->obd_recovery_task_lock); - LASSERT(obd->obd_recovering); - cfs_list_for_each(tmp, &obd->obd_req_replay_queue) { - struct ptlrpc_request *reqiter = - cfs_list_entry(tmp, struct ptlrpc_request, rq_list); - - if (lustre_msg_get_transno(reqiter->rq_reqmsg) > transno) { - cfs_list_add_tail(&req->rq_list, &reqiter->rq_list); - inserted = 1; - break; - } + 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)) { @@ -2327,23 +2354,21 @@ int target_queue_recovery_request(struct ptlrpc_request *req, RETURN(0); } } - +added: if (!inserted) - cfs_list_add_tail(&req->rq_list, &obd->obd_req_replay_queue); + 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); } -EXPORT_SYMBOL(target_queue_recovery_request); int target_handle_ping(struct ptlrpc_request *req) { obd_ping(req->rq_svc_thread->t_env, req->rq_export); return req_capsule_server_pack(&req->rq_pill); } -EXPORT_SYMBOL(target_handle_ping); void target_committed_to_req(struct ptlrpc_request *req) { @@ -2360,7 +2385,6 @@ void target_committed_to_req(struct ptlrpc_request *req) CDEBUG(D_INFO, "last_committed "LPU64", transno "LPU64", xid "LPU64"\n", exp->exp_last_committed, req->rq_transno, req->rq_xid); } -EXPORT_SYMBOL(target_committed_to_req); #endif /* HAVE_SERVER_SUPPORT */ @@ -2391,9 +2415,9 @@ int target_pack_pool_reply(struct ptlrpc_request *req) RETURN(0); } -EXPORT_SYMBOL(target_pack_pool_reply); -int target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id) +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"); @@ -2434,20 +2458,20 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) } /* must be an export if locks saved */ - LASSERT (req->rq_export != NULL); + 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 (cfs_list_empty(&rs->rs_obd_list)); - LASSERT (cfs_list_empty(&rs->rs_exp_list)); + 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); + exp = class_export_get(req->rq_export); /* disable reply scheduling while I'm setting up */ rs->rs_scheduled = 1; @@ -2462,13 +2486,13 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) rs->rs_transno, exp->exp_last_committed); if (rs->rs_transno > exp->exp_last_committed) { /* not committed already */ - cfs_list_add_tail(&rs->rs_obd_list, + list_add_tail(&rs->rs_obd_list, &exp->exp_uncommitted_replies); } spin_unlock(&exp->exp_uncommitted_replies_lock); spin_lock(&exp->exp_lock); - cfs_list_add_tail(&rs->rs_exp_list, &exp->exp_outstanding_replies); + 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); @@ -2490,19 +2514,18 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id) spin_lock(&rs->rs_lock); if (rs->rs_transno <= exp->exp_last_committed || (!rs->rs_on_net && !rs->rs_no_ack) || - cfs_list_empty(&rs->rs_exp_list) || /* completed already */ - cfs_list_empty(&rs->rs_obd_list)) { + 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 { - cfs_list_add(&rs->rs_list, &svcpt->scp_rep_active); + 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; } -EXPORT_SYMBOL(target_send_reply); ldlm_mode_t lck_compat_array[] = { [LCK_EX] = LCK_COMPAT_EX, @@ -2525,6 +2548,7 @@ int ldlm_error2errno(ldlm_error_t error) switch (error) { case ELDLM_OK: + case ELDLM_LOCK_MATCHED: result = 0; break; case ELDLM_LOCK_CHANGED: @@ -2591,18 +2615,17 @@ ldlm_error_t ldlm_errno2error(int err_no) } return error; } -EXPORT_SYMBOL(ldlm_errno2error); #if LUSTRE_TRACKS_LOCK_EXP_REFS void ldlm_dump_export_locks(struct obd_export *exp) { spin_lock(&exp->exp_locks_list_guard); - if (!cfs_list_empty(&exp->exp_locks_list)) { + 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); - cfs_list_for_each_entry(lock, &exp->exp_locks_list, + list_for_each_entry(lock, &exp->exp_locks_list, l_exp_refs_link) LDLM_ERROR(lock, "lock:"); } @@ -2620,9 +2643,13 @@ static int target_bulk_timeout(void *data) RETURN(1); } -static inline char *bulk2type(struct ptlrpc_bulk_desc *desc) +static inline const char *bulk2type(struct ptlrpc_request *req) { - return desc->bd_type == BULK_GET_SINK ? "GET" : "PUT"; + 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, @@ -2630,6 +2657,7 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, { struct ptlrpc_request *req = desc->bd_req; time_t start = cfs_time_current_sec(); + time_t deadline; int rc = 0; ENTRY; @@ -2639,7 +2667,7 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, *lwi = LWI_INTR(NULL, NULL); rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq, !atomic_read(&exp->exp_obd-> - obd_evict_inprogress), + obd_evict_inprogress), lwi); } @@ -2648,7 +2676,7 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, exp->exp_conn_cnt > lustre_msg_get_conn_cnt(req->rq_reqmsg)) { rc = -ENOTCONN; } else { - if (desc->bd_type == BULK_PUT_SINK) + if (req->rq_bulk_read) rc = sptlrpc_svc_wrap_bulk(req, desc); if (rc == 0) rc = ptlrpc_start_bulk_transfer(desc); @@ -2656,7 +2684,7 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, if (rc < 0) { DEBUG_REQ(D_ERROR, req, "bulk %s failed: rc %d", - bulk2type(desc), rc); + bulk2type(req), rc); RETURN(rc); } @@ -2665,10 +2693,16 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *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 { - long timeoutl = req->rq_deadline - cfs_time_current_sec(); + long timeoutl = deadline - cfs_time_current_sec(); cfs_duration_t timeout = timeoutl <= 0 ? CFS_TICK : cfs_time_seconds(timeoutl); + time_t rq_deadline; *lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1), target_bulk_timeout, desc); @@ -2679,37 +2713,46 @@ int target_bulk_io(struct obd_export *exp, struct ptlrpc_bulk_desc *desc, lustre_msg_get_conn_cnt(req->rq_reqmsg), lwi); LASSERT(rc == 0 || rc == -ETIMEDOUT); - /* Wait again if we changed deadline. */ + /* Wait again if we changed rq_deadline. */ + rq_deadline = ACCESS_ONCE(req->rq_deadline); + deadline = start + bulk_timeout; + if (deadline > rq_deadline) + deadline = rq_deadline; } while ((rc == -ETIMEDOUT) && - (req->rq_deadline > cfs_time_current_sec())); + (deadline > cfs_time_current_sec())); if (rc == -ETIMEDOUT) { DEBUG_REQ(D_ERROR, req, "timeout on bulk %s after %ld%+lds", - bulk2type(desc), req->rq_deadline - start, - cfs_time_current_sec() - req->rq_deadline); + bulk2type(req), deadline - start, + cfs_time_current_sec() - deadline); ptlrpc_abort_bulk(desc); } else if (exp->exp_failed) { DEBUG_REQ(D_ERROR, req, "Eviction on bulk %s", - bulk2type(desc)); + 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(desc)); + bulk2type(req)); /* We don't reply anyway. */ rc = -ETIMEDOUT; ptlrpc_abort_bulk(desc); - } else if (desc->bd_failure || - desc->bd_nob_transferred != desc->bd_nob) { - DEBUG_REQ(D_ERROR, req, "%s bulk %s %d(%d)", - desc->bd_failure ? "network error on" : "truncated", - bulk2type(desc), desc->bd_nob_transferred, - desc->bd_nob); - /* XXX Should this be a different errno? */ + } 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 (desc->bd_type == BULK_GET_SINK) { - rc = sptlrpc_svc_unwrap_bulk(req, desc); + } 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);