X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmgc%2Fmgc_request.c;h=411909f10c72557c052a224f6464369c0f9ef608;hb=7bf1d7c6cb7d0a7231b3fdcb9e3d3ec3129fb427;hp=9beed2936d930beb3a6ac45c71fd766cae8f19e6;hpb=35a8ed2b2007d89c1f125f01f155232e7f511e98;p=fs%2Flustre-release.git diff --git a/lustre/mgc/mgc_request.c b/lustre/mgc/mgc_request.c index 9beed29..411909f 100644 --- a/lustre/mgc/mgc_request.c +++ b/lustre/mgc/mgc_request.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, Whamcloud, Inc. + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -117,7 +117,7 @@ int mgc_logname2resid(char *logname, struct ldlm_res_id *res_id, int type) /********************** config llog list **********************/ static CFS_LIST_HEAD(config_llog_list); -static cfs_spinlock_t config_list_lock = CFS_SPIN_LOCK_UNLOCKED; +static DEFINE_SPINLOCK(config_list_lock); /* Take a reference to a config log */ static int config_log_get(struct config_llog_data *cld) @@ -142,7 +142,7 @@ static void config_log_put(struct config_llog_data *cld) /* spinlock to make sure no item with 0 refcount in the list */ if (cfs_atomic_dec_and_lock(&cld->cld_refcount, &config_list_lock)) { cfs_list_del(&cld->cld_list_chain); - cfs_spin_unlock(&config_list_lock); + spin_unlock(&config_list_lock); CDEBUG(D_MGC, "dropping config log %s\n", cld->cld_logname); @@ -173,7 +173,7 @@ struct config_llog_data *config_log_find(char *logname, LASSERT(logname != NULL); instance = cfg ? cfg->cfg_instance : NULL; - cfs_spin_lock(&config_list_lock); + spin_lock(&config_list_lock); cfs_list_for_each_entry(cld, &config_llog_list, cld_list_chain) { /* check if instance equals */ if (instance != cld->cld_cfg.cfg_instance) @@ -189,8 +189,8 @@ struct config_llog_data *config_log_find(char *logname, cfs_atomic_inc(&found->cld_refcount); LASSERT(found->cld_stopping == 0 || cld_is_sptlrpc(found) == 0); } - cfs_spin_unlock(&config_list_lock); - RETURN(found); + spin_unlock(&config_list_lock); + RETURN(found); } static @@ -214,7 +214,9 @@ struct config_llog_data *do_config_log_add(struct obd_device *obd, strcpy(cld->cld_logname, logname); if (cfg) cld->cld_cfg = *cfg; - cfs_mutex_init(&cld->cld_lock); + else + cld->cld_cfg.cfg_callback = class_config_llog_handler; + mutex_init(&cld->cld_lock); cld->cld_cfg.cfg_last_idx = 0; cld->cld_cfg.cfg_flags = 0; cld->cld_cfg.cfg_sb = sb; @@ -231,9 +233,9 @@ struct config_llog_data *do_config_log_add(struct obd_device *obd, rc = mgc_logname2resid(logname, &cld->cld_resid, type); - cfs_spin_lock(&config_list_lock); - cfs_list_add(&cld->cld_list_chain, &config_llog_list); - cfs_spin_unlock(&config_list_lock); + spin_lock(&config_list_lock); + cfs_list_add(&cld->cld_list_chain, &config_llog_list); + spin_unlock(&config_list_lock); if (rc) { config_log_put(cld); @@ -242,7 +244,7 @@ struct config_llog_data *do_config_log_add(struct obd_device *obd, if (cld_is_sptlrpc(cld)) { rc = mgc_process_log(obd, cld); - if (rc) + if (rc && rc != -ENOENT) CERROR("failed processing sptlrpc log: %d\n", rc); } @@ -259,14 +261,18 @@ static struct config_llog_data *config_recover_log_add(struct obd_device *obd, struct config_llog_data *cld; char logname[32]; - if ((lsi->lsi_flags & LSI_SERVER) && !IS_MDT(lsi->lsi_ldd)) + if (IS_OST(lsi)) return NULL; + /* for osp-on-ost, see lustre_start_osp() */ + if (IS_MDT(lsi) && lcfg.cfg_instance) + return NULL; + /* we have to use different llog for clients and mdts for cmd * where only clients are notified if one of cmd server restarts */ LASSERT(strlen(fsname) < sizeof(logname) / 2); strcpy(logname, fsname); - if (lsi->lsi_flags & LSI_SERVER) { /* mdt */ + if (IS_SERVER(lsi)) { /* mdt */ LASSERT(lcfg.cfg_instance == NULL); lcfg.cfg_instance = sb; strcat(logname, "-mdtir"); @@ -345,7 +351,7 @@ static int config_log_add(struct obd_device *obd, char *logname, RETURN(0); } -CFS_DEFINE_MUTEX(llog_process_lock); +DEFINE_MUTEX(llog_process_lock); /** Stop watching for updates on this log. */ @@ -361,7 +367,7 @@ static int config_log_end(char *logname, struct config_llog_instance *cfg) if (cld == NULL) RETURN(-ENOENT); - cfs_mutex_lock(&cld->cld_lock); + mutex_lock(&cld->cld_lock); /* * if cld_stopping is set, it means we didn't start the log thus * not owning the start ref. this can happen after previous umount: @@ -370,7 +376,7 @@ static int config_log_end(char *logname, struct config_llog_instance *cfg) * calling start_log. */ if (unlikely(cld->cld_stopping)) { - cfs_mutex_unlock(&cld->cld_lock); + mutex_unlock(&cld->cld_lock); /* drop the ref from the find */ config_log_put(cld); RETURN(rc); @@ -380,19 +386,19 @@ static int config_log_end(char *logname, struct config_llog_instance *cfg) cld_recover = cld->cld_recover; cld->cld_recover = NULL; - cfs_mutex_unlock(&cld->cld_lock); + mutex_unlock(&cld->cld_lock); - if (cld_recover) { - cfs_mutex_lock(&cld_recover->cld_lock); - cld_recover->cld_stopping = 1; - cfs_mutex_unlock(&cld_recover->cld_lock); - config_log_put(cld_recover); - } + if (cld_recover) { + mutex_lock(&cld_recover->cld_lock); + cld_recover->cld_stopping = 1; + mutex_unlock(&cld_recover->cld_lock); + config_log_put(cld_recover); + } - cfs_spin_lock(&config_list_lock); - cld_sptlrpc = cld->cld_sptlrpc; - cld->cld_sptlrpc = NULL; - cfs_spin_unlock(&config_list_lock); + spin_lock(&config_list_lock); + cld_sptlrpc = cld->cld_sptlrpc; + cld->cld_sptlrpc = NULL; + spin_unlock(&config_list_lock); if (cld_sptlrpc) config_log_put(cld_sptlrpc); @@ -418,21 +424,21 @@ int lprocfs_mgc_rd_ir_state(char *page, char **start, off_t off, ENTRY; rc = snprintf(page, count, "imperative_recovery: %s\n", - OCD_HAS_FLAG(ocd, IMP_RECOV) ? "ON" : "OFF"); + OCD_HAS_FLAG(ocd, IMP_RECOV) ? "ENABLED" : "DISABLED"); rc += snprintf(page + rc, count - rc, "client_state:\n"); - cfs_spin_lock(&config_list_lock); - cfs_list_for_each_entry(cld, &config_llog_list, cld_list_chain) { - if (cld->cld_recover == NULL) - continue; - rc += snprintf(page + rc, count - rc, - " - { client: %s, nidtbl_version: %u }\n", - cld->cld_logname, - cld->cld_recover->cld_cfg.cfg_last_idx); - } - cfs_spin_unlock(&config_list_lock); - - RETURN(rc); + spin_lock(&config_list_lock); + cfs_list_for_each_entry(cld, &config_llog_list, cld_list_chain) { + if (cld->cld_recover == NULL) + continue; + rc += snprintf(page + rc, count - rc, + " - { client: %s, nidtbl_version: %u }\n", + cld->cld_logname, + cld->cld_recover->cld_cfg.cfg_last_idx); + } + spin_unlock(&config_list_lock); + + RETURN(rc); } /* reenqueue any lost locks */ @@ -442,7 +448,7 @@ int lprocfs_mgc_rd_ir_state(char *page, char **start, off_t off, #define RQ_STOP 0x8 static int rq_state = 0; static cfs_waitq_t rq_waitq; -static CFS_DECLARE_COMPLETION(rq_exit); +static DECLARE_COMPLETION(rq_exit); static void do_requeue(struct config_llog_data *cld) { @@ -452,7 +458,7 @@ static void do_requeue(struct config_llog_data *cld) /* Do not run mgc_process_log on a disconnected export or an export which is being disconnected. Take the client semaphore to make the check non-racy. */ - cfs_down_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem); + down_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem); if (cld->cld_mgcexp->exp_obd->u.cli.cl_conn_count != 0) { CDEBUG(D_MGC, "updating log %s\n", cld->cld_logname); mgc_process_log(cld->cld_mgcexp->exp_obd, cld); @@ -460,7 +466,7 @@ static void do_requeue(struct config_llog_data *cld) CDEBUG(D_MGC, "disconnecting, won't update log %s\n", cld->cld_logname); } - cfs_up_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem); + up_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem); EXIT; } @@ -483,18 +489,18 @@ static int mgc_requeue_thread(void *data) CDEBUG(D_MGC, "Starting requeue thread\n"); /* Keep trying failed locks periodically */ - cfs_spin_lock(&config_list_lock); - rq_state |= RQ_RUNNING; - while (1) { - struct l_wait_info lwi; - struct config_llog_data *cld, *cld_prev; - int rand = cfs_rand() & MGC_TIMEOUT_RAND_CENTISEC; - int stopped = !!(rq_state & RQ_STOP); - int to; - - /* Any new or requeued lostlocks will change the state */ - rq_state &= ~(RQ_NOW | RQ_LATER); - cfs_spin_unlock(&config_list_lock); + spin_lock(&config_list_lock); + rq_state |= RQ_RUNNING; + while (1) { + struct l_wait_info lwi; + struct config_llog_data *cld, *cld_prev; + int rand = cfs_rand() & MGC_TIMEOUT_RAND_CENTISEC; + int stopped = !!(rq_state & RQ_STOP); + int to; + + /* Any new or requeued lostlocks will change the state */ + rq_state &= ~(RQ_NOW | RQ_LATER); + spin_unlock(&config_list_lock); /* Always wait a few seconds to allow the server who caused the lock revocation to finish its setup, plus some @@ -514,13 +520,13 @@ static int mgc_requeue_thread(void *data) */ cld_prev = NULL; - cfs_spin_lock(&config_list_lock); - cfs_list_for_each_entry(cld, &config_llog_list, - cld_list_chain) { - if (!cld->cld_lostlock) - continue; + spin_lock(&config_list_lock); + cfs_list_for_each_entry(cld, &config_llog_list, + cld_list_chain) { + if (!cld->cld_lostlock) + continue; - cfs_spin_unlock(&config_list_lock); + spin_unlock(&config_list_lock); LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0); @@ -534,33 +540,33 @@ static int mgc_requeue_thread(void *data) if (likely(!stopped)) do_requeue(cld); - cfs_spin_lock(&config_list_lock); - } - cfs_spin_unlock(&config_list_lock); - if (cld_prev) - config_log_put(cld_prev); - - /* break after scanning the list so that we can drop - * refcount to losing lock clds */ - if (unlikely(stopped)) { - cfs_spin_lock(&config_list_lock); - break; - } - - /* Wait a bit to see if anyone else needs a requeue */ - lwi = (struct l_wait_info) { 0 }; - l_wait_event(rq_waitq, rq_state & (RQ_NOW | RQ_STOP), - &lwi); - cfs_spin_lock(&config_list_lock); - } - /* spinlock and while guarantee RQ_NOW and RQ_LATER are not set */ - rq_state &= ~RQ_RUNNING; - cfs_spin_unlock(&config_list_lock); - - cfs_complete(&rq_exit); - - CDEBUG(D_MGC, "Ending requeue thread\n"); - RETURN(rc); + spin_lock(&config_list_lock); + } + spin_unlock(&config_list_lock); + if (cld_prev) + config_log_put(cld_prev); + + /* break after scanning the list so that we can drop + * refcount to losing lock clds */ + if (unlikely(stopped)) { + spin_lock(&config_list_lock); + break; + } + + /* Wait a bit to see if anyone else needs a requeue */ + lwi = (struct l_wait_info) { 0 }; + l_wait_event(rq_waitq, rq_state & (RQ_NOW | RQ_STOP), + &lwi); + spin_lock(&config_list_lock); + } + /* spinlock and while guarantee RQ_NOW and RQ_LATER are not set */ + rq_state &= ~RQ_RUNNING; + spin_unlock(&config_list_lock); + + complete(&rq_exit); + + CDEBUG(D_MGC, "Ending requeue thread\n"); + RETURN(rc); } /* Add a cld to the list to requeue. Start the requeue thread if needed. @@ -574,28 +580,28 @@ static void mgc_requeue_add(struct config_llog_data *cld) cld->cld_stopping, rq_state); LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0); - cfs_mutex_lock(&cld->cld_lock); - if (cld->cld_stopping || cld->cld_lostlock) { - cfs_mutex_unlock(&cld->cld_lock); - RETURN_EXIT; - } - /* this refcount will be released in mgc_requeue_thread. */ - config_log_get(cld); - cld->cld_lostlock = 1; - cfs_mutex_unlock(&cld->cld_lock); - - /* Hold lock for rq_state */ - cfs_spin_lock(&config_list_lock); - if (rq_state & RQ_STOP) { - cfs_spin_unlock(&config_list_lock); - cld->cld_lostlock = 0; - config_log_put(cld); - } else { - rq_state |= RQ_NOW; - cfs_spin_unlock(&config_list_lock); - cfs_waitq_signal(&rq_waitq); - } - EXIT; + mutex_lock(&cld->cld_lock); + if (cld->cld_stopping || cld->cld_lostlock) { + mutex_unlock(&cld->cld_lock); + RETURN_EXIT; + } + /* this refcount will be released in mgc_requeue_thread. */ + config_log_get(cld); + cld->cld_lostlock = 1; + mutex_unlock(&cld->cld_lock); + + /* Hold lock for rq_state */ + spin_lock(&config_list_lock); + if (rq_state & RQ_STOP) { + spin_unlock(&config_list_lock); + cld->cld_lostlock = 0; + config_log_put(cld); + } else { + rq_state |= RQ_NOW; + spin_unlock(&config_list_lock); + cfs_waitq_signal(&rq_waitq); + } + EXIT; } /********************** class fns **********************/ @@ -615,15 +621,15 @@ static int mgc_fs_setup(struct obd_device *obd, struct super_block *sb, LASSERT(lsi->lsi_srv_mnt == mnt); /* The mgc fs exclusion sem. Only one fs can be setup at a time. */ - cfs_down(&cli->cl_mgc_sem); + down(&cli->cl_mgc_sem); cfs_cleanup_group_info(); - obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd)); + obd->obd_fsops = fsfilt_get_ops(lsi->lsi_fstype); if (IS_ERR(obd->obd_fsops)) { - cfs_up(&cli->cl_mgc_sem); - CERROR("No fstype %s rc=%ld\n", MT_STR(lsi->lsi_ldd), - PTR_ERR(obd->obd_fsops)); + up(&cli->cl_mgc_sem); + CERROR("%s: No fstype %s: rc = %ld\n", lsi->lsi_fstype, + obd->obd_name, PTR_ERR(obd->obd_fsops)); RETURN(PTR_ERR(obd->obd_fsops)); } @@ -664,7 +670,7 @@ err_ops: fsfilt_put_ops(obd->obd_fsops); obd->obd_fsops = NULL; cli->cl_mgc_vfsmnt = NULL; - cfs_up(&cli->cl_mgc_sem); + up(&cli->cl_mgc_sem); RETURN(err); } @@ -689,7 +695,7 @@ static int mgc_fs_cleanup(struct obd_device *obd) if (obd->obd_fsops) fsfilt_put_ops(obd->obd_fsops); - cfs_up(&cli->cl_mgc_sem); + up(&cli->cl_mgc_sem); RETURN(rc); } @@ -707,14 +713,14 @@ static int mgc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) if (cfs_atomic_dec_and_test(&mgc_count)) { int running; /* stop requeue thread */ - cfs_spin_lock(&config_list_lock); - running = rq_state & RQ_RUNNING; - if (running) - rq_state |= RQ_STOP; - cfs_spin_unlock(&config_list_lock); - if (running) { - cfs_waitq_signal(&rq_waitq); - cfs_wait_for_completion(&rq_exit); + spin_lock(&config_list_lock); + running = rq_state & RQ_RUNNING; + if (running) + rq_state |= RQ_STOP; + spin_unlock(&config_list_lock); + if (running) { + cfs_waitq_signal(&rq_waitq); + wait_for_completion(&rq_exit); } } obd_cleanup_client_import(obd); @@ -899,9 +905,9 @@ static int mgc_set_mgs_param(struct obd_export *exp, /* Take a config lock so we can get cancel notifications */ static int mgc_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm, __u32 type, ldlm_policy_data_t *policy, __u32 mode, - int *flags, void *bl_cb, void *cp_cb, void *gl_cb, - void *data, __u32 lvb_len, void *lvb_swabber, - struct lustre_handle *lockh) + __u64 *flags, void *bl_cb, void *cp_cb, void *gl_cb, + void *data, __u32 lvb_len, void *lvb_swabber, + struct lustre_handle *lockh) { struct config_llog_data *cld = (struct config_llog_data *)data; struct ldlm_enqueue_info einfo = { type, mode, mgc_blocking_ast, @@ -921,18 +927,20 @@ static int mgc_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm, LDLM_ENQUEUE); if (req == NULL) RETURN(-ENOMEM); + + req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, 0); ptlrpc_request_set_replen(req); /* check if this is server or client */ if (cld->cld_cfg.cfg_sb) { struct lustre_sb_info *lsi = s2lsi(cld->cld_cfg.cfg_sb); - if (lsi && (lsi->lsi_flags & LSI_SERVER)) + if (lsi && IS_SERVER(lsi)) short_limit = 1; } /* Limit how long we will wait for the enqueue to complete */ req->rq_delay_limit = short_limit ? 5 : MGC_ENQUEUE_LIMIT; rc = ldlm_cli_enqueue(exp, &req, &einfo, &cld->cld_resid, NULL, flags, - NULL, 0, lockh, 0); + NULL, 0, LVB_T_NONE, lockh, 0); /* A failed enqueue should still call the mgc_blocking_ast, where it will be requeued if needed ("grant failed"). */ ptlrpc_req_finished(req); @@ -951,13 +959,13 @@ static int mgc_cancel(struct obd_export *exp, struct lov_stripe_md *md, static void mgc_notify_active(struct obd_device *unused) { - /* wakeup mgc_requeue_thread to requeue mgc lock */ - cfs_spin_lock(&config_list_lock); - rq_state |= RQ_NOW; - cfs_spin_unlock(&config_list_lock); - cfs_waitq_signal(&rq_waitq); + /* wakeup mgc_requeue_thread to requeue mgc lock */ + spin_lock(&config_list_lock); + rq_state |= RQ_NOW; + spin_unlock(&config_list_lock); + cfs_waitq_signal(&rq_waitq); - /* TODO: Help the MGS rebuild nidtbl. -jay */ + /* TODO: Help the MGS rebuild nidtbl. -jay */ } /* Send target_reg message to MGS */ @@ -1150,12 +1158,12 @@ static int mgc_import_event(struct obd_device *obd, ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY); break; } - case IMP_EVENT_ACTIVE: - LCONSOLE_WARN("%s: Reactivating import\n", obd->obd_name); - /* Clearing obd_no_recov allows us to continue pinging */ - obd->obd_no_recov = 0; - mgc_notify_active(obd); - break; + case IMP_EVENT_ACTIVE: + CDEBUG(D_INFO, "%s: Reactivating import\n", obd->obd_name); + /* Clearing obd_no_recov allows us to continue pinging */ + obd->obd_no_recov = 0; + mgc_notify_active(obd); + break; case IMP_EVENT_OCD: break; case IMP_EVENT_DEACTIVATE: @@ -1177,50 +1185,47 @@ static int mgc_llog_init(struct obd_device *obd, struct obd_llog_group *olg, LASSERT(olg == &obd->obd_olg); - rc = llog_setup(obd, olg, LLOG_CONFIG_ORIG_CTXT, tgt, 0, NULL, - &llog_lvfs_ops); - if (rc) - RETURN(rc); +#ifdef HAVE_LDISKFS_OSD + rc = llog_setup(NULL, obd, olg, LLOG_CONFIG_ORIG_CTXT, tgt, + &llog_lvfs_ops); + if (rc) + RETURN(rc); +#endif - rc = llog_setup(obd, olg, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL, - &llog_client_ops); - if (rc == 0) { - ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT); - if (!ctxt) { - ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); - if (ctxt) - llog_cleanup(ctxt); - RETURN(-ENODEV); - } - llog_initiator_connect(ctxt); - llog_ctxt_put(ctxt); - } else { - ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); - if (ctxt) - llog_cleanup(ctxt); - } + rc = llog_setup(NULL, obd, olg, LLOG_CONFIG_REPL_CTXT, tgt, + &llog_client_ops); + if (rc) + GOTO(out, rc); + + ctxt = llog_group_get_ctxt(olg, LLOG_CONFIG_REPL_CTXT); + if (!ctxt) + GOTO(out, rc = -ENODEV); + llog_initiator_connect(ctxt); + llog_ctxt_put(ctxt); + + RETURN(0); +out: + ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); + if (ctxt) + llog_cleanup(NULL, ctxt); RETURN(rc); } static int mgc_llog_finish(struct obd_device *obd, int count) { - struct llog_ctxt *ctxt; - int rc = 0, rc2 = 0; - ENTRY; - - ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT); - if (ctxt) - rc = llog_cleanup(ctxt); + struct llog_ctxt *ctxt; - ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); - if (ctxt) - rc2 = llog_cleanup(ctxt); + ENTRY; - if (!rc) - rc = rc2; + ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT); + if (ctxt) + llog_cleanup(NULL, ctxt); - RETURN(rc); + ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); + if (ctxt) + llog_cleanup(NULL, ctxt); + RETURN(0); } enum { @@ -1229,9 +1234,9 @@ enum { }; static int mgc_apply_recover_logs(struct obd_device *mgc, - struct config_llog_data *cld, - __u64 max_version, - void *data, int datalen, int need_swab) + struct config_llog_data *cld, + __u64 max_version, + void *data, int datalen, bool mne_swab) { struct config_llog_instance *cfg = &cld->cld_cfg; struct lustre_sb_info *lsi = s2lsi(cfg->cfg_sb); @@ -1254,11 +1259,16 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, if (inst == NULL) RETURN(-ENOMEM); - if (!(lsi->lsi_flags & LSI_SERVER)) { + if (!IS_SERVER(lsi)) { pos = sprintf(inst, "%p", cfg->cfg_instance); } else { - LASSERT(IS_MDT(lsi->lsi_ldd)); - pos = sprintf(inst, "MDT%04x", lsi->lsi_ldd->ldd_svindex); + LASSERT(IS_MDT(lsi)); + rc = server_name2svname(lsi->lsi_svname, inst, NULL); + if (rc) { + OBD_FREE(inst, CFS_PAGE_SIZE); + RETURN(-EINVAL); + } + pos = strlen(inst); } ++pos; @@ -1292,11 +1302,16 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, if (datalen < entry_len) /* must have entry_len at least */ break; - if (need_swab) - lustre_swab_mgs_nidtbl_entry(entry); - LASSERT(entry->mne_length <= CFS_PAGE_SIZE); - if (entry->mne_length < entry_len) - break; + /* Keep this swab for normal mixed endian handling. LU-1644 */ + if (mne_swab) + lustre_swab_mgs_nidtbl_entry(entry); + if (entry->mne_length > CFS_PAGE_SIZE) { + CERROR("MNE too large (%u)\n", entry->mne_length); + break; + } + + if (entry->mne_length < entry_len) + break; off += entry->mne_length; datalen -= entry->mne_length; @@ -1349,7 +1364,7 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, if (obd == NULL) { CDEBUG(D_INFO, "mgc %s: cannot find obdname %s\n", mgc->obd_name, obdname); - + rc = 0; /* this is a safe race, when the ost is starting up...*/ continue; } @@ -1360,11 +1375,20 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, pos += sprintf(params, "%s.import=%s", cname, "connection="); uuid = buf + pos; + down_read(&obd->u.cli.cl_sem); + if (obd->u.cli.cl_import == NULL) { + /* client does not connect to the OST yet */ + up_read(&obd->u.cli.cl_sem); + rc = 0; + continue; + } + /* TODO: iterate all nids to find one */ /* find uuid by nid */ rc = client_import_find_conn(obd->u.cli.cl_import, entry->u.nids[0], (struct obd_uuid *)uuid); + up_read(&obd->u.cli.cl_sem); if (rc < 0) { CERROR("mgc: cannot find uuid by nid %s\n", libcfs_nid2str(entry->u.nids[0])); @@ -1418,6 +1442,7 @@ static int mgc_process_recover_log(struct obd_device *obd, cfs_page_t **pages; int nrpages; bool eof = true; + bool mne_swab = false; int i; int ealen; int rc; @@ -1445,7 +1470,7 @@ static int mgc_process_recover_log(struct obd_device *obd, again: LASSERT(cld_is_recover(cld)); - LASSERT(cfs_mutex_is_locked(&cld->cld_lock)); + LASSERT(mutex_is_locked(&cld->cld_lock)); req = ptlrpc_request_alloc(class_exp2cliimp(cld->cld_mgcexp), &RQF_MGS_CONFIG_READ); if (req == NULL) @@ -1472,7 +1497,7 @@ again: GOTO(out, rc = -ENOMEM); for (i = 0; i < nrpages; i++) - ptlrpc_prep_bulk_page(desc, pages[i], 0, CFS_PAGE_SIZE); + ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, CFS_PAGE_SIZE); ptlrpc_request_set_replen(req); rc = ptlrpc_queue_wait(req); @@ -1504,14 +1529,24 @@ again: GOTO(out, rc); } + mne_swab = !!ptlrpc_rep_need_swab(req); +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0) + /* This import flag means the server did an extra swab of IR MNE + * records (fixed in LU-1252), reverse it here if needed. LU-1644 */ + if (unlikely(req->rq_import->imp_need_mne_swab)) + mne_swab = !mne_swab; +#else +#warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab" +#endif + for (i = 0; i < nrpages && ealen > 0; i++) { int rc2; void *ptr; ptr = cfs_kmap(pages[i]); rc2 = mgc_apply_recover_logs(obd, cld, res->mcr_offset, ptr, - min_t(int, ealen, CFS_PAGE_SIZE), - ptlrpc_rep_need_swab(req)); + min_t(int, ealen, CFS_PAGE_SIZE), + mne_swab); cfs_kunmap(pages[i]); if (rc2 < 0) { CWARN("Process recover log %s error %d\n", @@ -1540,47 +1575,33 @@ out: return rc; } -/* identical to mgs_log_is_empty */ -static int mgc_llog_is_empty(struct obd_device *obd, struct llog_ctxt *ctxt, - char *name) -{ - struct lvfs_run_ctxt saved; - struct llog_handle *llh; - int rc = 0; +#ifdef HAVE_LDISKFS_OSD - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - rc = llog_create(ctxt, &llh, NULL, name); - if (rc == 0) { - llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL); - rc = llog_get_size(llh); - llog_close(llh); - } - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - /* header is record 1 */ - return(rc <= 1); -} +/* + * XXX: mgc_copy_llog() does not support osd-based llogs yet + */ -static int mgc_copy_handler(struct llog_handle *llh, struct llog_rec_hdr *rec, - void *data) +/* identical to mgs_log_is_empty */ +static int mgc_llog_is_empty(struct obd_device *obd, struct llog_ctxt *ctxt, + char *name) { - struct llog_rec_hdr local_rec = *rec; - struct llog_handle *local_llh = (struct llog_handle *)data; - char *cfg_buf = (char*) (rec + 1); - struct lustre_cfg *lcfg; - int rc = 0; - ENTRY; - - /* Append all records */ - local_rec.lrh_len -= sizeof(*rec) + sizeof(struct llog_rec_tail); - rc = llog_write_rec(local_llh, &local_rec, NULL, 0, - (void *)cfg_buf, -1); - - lcfg = (struct lustre_cfg *)cfg_buf; - CDEBUG(D_INFO, "idx=%d, rc=%d, len=%d, cmd %x %s %s\n", - rec->lrh_index, rc, rec->lrh_len, lcfg->lcfg_command, - lustre_cfg_string(lcfg, 0), lustre_cfg_string(lcfg, 1)); - - RETURN(rc); + struct lvfs_run_ctxt saved; + struct llog_handle *llh; + int rc = 0; + + push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + rc = llog_open(NULL, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS); + if (rc == 0) { + rc = llog_init_handle(NULL, llh, LLOG_F_IS_PLAIN, NULL); + if (rc == 0) + rc = llog_get_size(llh); + llog_close(NULL, llh); + } else if (rc == -ENOENT) { + rc = 0; + } + pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + /* header is record 1 */ + return (rc <= 1); } /* Copy a remote log locally */ @@ -1602,54 +1623,54 @@ static int mgc_copy_llog(struct obd_device *obd, struct llog_ctxt *rctxt, sprintf(temp_log, "%sT", logname); /* Make sure there's no old temp log */ - rc = llog_create(lctxt, &local_llh, NULL, temp_log); - if (rc) - GOTO(out, rc); - rc = llog_init_handle(local_llh, LLOG_F_IS_PLAIN, NULL); - if (rc) - GOTO(out, rc); - rc = llog_destroy(local_llh); - llog_free_handle(local_llh); - if (rc) - GOTO(out, rc); - - /* open local log */ - rc = llog_create(lctxt, &local_llh, NULL, temp_log); - if (rc) - GOTO(out, rc); - - /* set the log header uuid for fun */ - OBD_ALLOC_PTR(uuid); - obd_str2uuid(uuid, logname); - rc = llog_init_handle(local_llh, LLOG_F_IS_PLAIN, uuid); - OBD_FREE_PTR(uuid); - if (rc) - GOTO(out_closel, rc); - - /* open remote log */ - rc = llog_create(rctxt, &remote_llh, NULL, logname); - if (rc) - GOTO(out_closel, rc); - rc = llog_init_handle(remote_llh, LLOG_F_IS_PLAIN, NULL); - if (rc) - GOTO(out_closer, rc); - - /* Copy remote log */ - rc = llog_process(remote_llh, mgc_copy_handler,(void *)local_llh, NULL); + rc = llog_erase(NULL, lctxt, NULL, temp_log); + if (rc < 0 && rc != -ENOENT) + GOTO(out, rc); + + /* open local log */ + rc = llog_open_create(NULL, lctxt, &local_llh, NULL, temp_log); + if (rc) + GOTO(out, rc); + + /* set the log header uuid for fun */ + OBD_ALLOC_PTR(uuid); + obd_str2uuid(uuid, logname); + rc = llog_init_handle(NULL, local_llh, LLOG_F_IS_PLAIN, uuid); + OBD_FREE_PTR(uuid); + if (rc) + GOTO(out_closel, rc); + + /* open remote log */ + rc = llog_open(NULL, rctxt, &remote_llh, NULL, logname, + LLOG_OPEN_EXISTS); + if (rc < 0) { + if (rc == -ENOENT) + rc = 0; + GOTO(out_closel, rc); + } + + rc = llog_init_handle(NULL, remote_llh, LLOG_F_IS_PLAIN, NULL); + if (rc) + GOTO(out_closer, rc); + + /* Copy remote log */ + rc = llog_process(NULL, remote_llh, llog_copy_handler, + (void *)local_llh, NULL); out_closer: - rc2 = llog_close(remote_llh); - if (!rc) - rc = rc2; + rc2 = llog_close(NULL, remote_llh); + if (!rc) + rc = rc2; out_closel: - rc2 = llog_close(local_llh); + rc2 = llog_close(NULL, local_llh); if (!rc) rc = rc2; /* We've copied the remote log to the local temp log, now replace the old local log with the temp log. */ - if (!rc) { + if (rc == 0) { struct client_obd *cli = &obd->u.cli; + LASSERT(cli); LASSERT(cli->cl_mgc_configs_dir); rc = lustre_rename(cli->cl_mgc_configs_dir, cli->cl_mgc_vfsmnt, @@ -1662,6 +1683,7 @@ out: OBD_FREE(temp_log, strlen(logname) + 1); RETURN(rc); } +#endif /* local_only means it cannot get remote llogs */ static int mgc_process_cfg_log(struct obd_device *mgc, @@ -1669,7 +1691,9 @@ static int mgc_process_cfg_log(struct obd_device *mgc, int local_only) { struct llog_ctxt *ctxt, *lctxt = NULL; +#ifdef HAVE_LDISKFS_OSD struct client_obd *cli = &mgc->u.cli; +#endif struct lvfs_run_ctxt *saved_ctxt; struct lustre_sb_info *lsi = NULL; int rc = 0, must_pop = 0; @@ -1678,7 +1702,7 @@ static int mgc_process_cfg_log(struct obd_device *mgc, ENTRY; LASSERT(cld); - LASSERT(cfs_mutex_is_locked(&cld->cld_lock)); + LASSERT(mutex_is_locked(&cld->cld_lock)); /* * local copy of sptlrpc log is controlled elsewhere, don't try to @@ -1702,11 +1726,15 @@ static int mgc_process_cfg_log(struct obd_device *mgc, lctxt = llog_get_context(mgc, LLOG_CONFIG_ORIG_CTXT); +#ifdef HAVE_LDISKFS_OSD + /* + * XXX: at the moment mgc_copy_llog() works with lvfs-based llogs + */ /* Copy the setup log locally if we can. Don't mess around if we're running an MGS though (logs are already local). */ - if (lctxt && lsi && (lsi->lsi_flags & LSI_SERVER) && + if (lctxt && lsi && IS_SERVER(lsi) && (lsi->lsi_srv_mnt == cli->cl_mgc_vfsmnt) && - !IS_MGS(lsi->lsi_ldd)) { + !IS_MGS(lsi) && lsi->lsi_srv_mnt) { push_ctxt(saved_ctxt, &mgc->obd_lvfs_ctxt, NULL); must_pop++; if (!local_only) @@ -1729,7 +1757,9 @@ static int mgc_process_cfg_log(struct obd_device *mgc, llog_ctxt_put(ctxt); ctxt = lctxt; lctxt = NULL; - } else if (local_only) { /* no local log at client side */ + } else +#endif + if (local_only) { /* no local log at client side */ GOTO(out_pop, rc = -EIO); } @@ -1741,7 +1771,8 @@ static int mgc_process_cfg_log(struct obd_device *mgc, /* logname and instance info should be the same, so use our copy of the instance for the update. The cfg_last_idx will be updated here. */ - rc = class_config_parse_llog(ctxt, cld->cld_logname, &cld->cld_cfg); + rc = class_config_parse_llog(NULL, ctxt, cld->cld_logname, + &cld->cld_cfg); EXIT; out_pop: @@ -1776,7 +1807,8 @@ out_pop: int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld) { struct lustre_handle lockh = { 0 }; - int rc = 0, rcl, flags = LDLM_FL_NO_LRU; + __u64 flags = LDLM_FL_NO_LRU; + int rc = 0, rcl; ENTRY; LASSERT(cld); @@ -1785,9 +1817,9 @@ int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld) sounds like badness. It actually might be fine, as long as we're not trying to update from the same log simultaneously (in which case we should use a per-log sem.) */ - cfs_mutex_lock(&cld->cld_lock); - if (cld->cld_stopping) { - cfs_mutex_unlock(&cld->cld_lock); + mutex_lock(&cld->cld_lock); + if (cld->cld_stopping) { + mutex_unlock(&cld->cld_lock); RETURN(0); } @@ -1827,7 +1859,7 @@ int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld) CDEBUG(D_MGC, "%s: configuration from log '%s' %sed (%d).\n", mgc->obd_name, cld->cld_logname, rc ? "fail" : "succeed", rc); - cfs_mutex_unlock(&cld->cld_lock); + mutex_unlock(&cld->cld_lock); /* Now drop the lock so MGS can revoke it */ if (!rcl) {