X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmgc%2Fmgc_request.c;h=b263c4b4cce1e83de0867a1e98e4df1ecd905f24;hp=089cfedd373b4103371f6c37da902e6493c8f3be;hb=45ce1044cd7b94621e1161cd23c600f8e1c18317;hpb=17f35d4948a0970fa76dc491037c0a8e8d95e813 diff --git a/lustre/mgc/mgc_request.c b/lustre/mgc/mgc_request.c index 089cfed..b263c4b 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, 2013, Intel Corporation. + * Copyright (c) 2011, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -42,12 +42,16 @@ #define D_MGC D_CONFIG /*|D_WARNING*/ #include -#include -#include +#include + +#include #include -#include +#include #include -#include +#include +#include +#include +#include #include "mgc_internal.h" @@ -77,8 +81,9 @@ static int mgc_name2resid(char *name, int len, struct ldlm_res_id *res_id, break; case CONFIG_T_RECOVER: case CONFIG_T_PARAMS: - resname = type; - break; + case CONFIG_T_NODEMAP: + resname = type; + break; default: LBUG(); } @@ -142,22 +147,24 @@ static void config_log_put(struct config_llog_data *cld) list_del(&cld->cld_list_chain); spin_unlock(&config_list_lock); - CDEBUG(D_MGC, "dropping config log %s\n", cld->cld_logname); + CDEBUG(D_MGC, "dropping config log %s\n", cld->cld_logname); - if (cld->cld_recover) - config_log_put(cld->cld_recover); - if (cld->cld_sptlrpc) - config_log_put(cld->cld_sptlrpc); + if (cld->cld_recover) + config_log_put(cld->cld_recover); + if (cld->cld_sptlrpc) + config_log_put(cld->cld_sptlrpc); if (cld->cld_params) config_log_put(cld->cld_params); - if (cld_is_sptlrpc(cld)) - sptlrpc_conf_log_stop(cld->cld_logname); + if (cld->cld_nodemap) + config_log_put(cld->cld_nodemap); + if (cld_is_sptlrpc(cld)) + sptlrpc_conf_log_stop(cld->cld_logname); - class_export_put(cld->cld_mgcexp); - OBD_FREE(cld, sizeof(*cld) + strlen(cld->cld_logname) + 1); - } + class_export_put(cld->cld_mgcexp); + OBD_FREE(cld, sizeof(*cld) + strlen(cld->cld_logname) + 1); + } - EXIT; + EXIT; } /* Find a config log by name */ @@ -195,21 +202,22 @@ struct config_llog_data *config_log_find(char *logname, static struct config_llog_data *do_config_log_add(struct obd_device *obd, - char *logname, - int type, - struct config_llog_instance *cfg, - struct super_block *sb) + char *logname, + int type, + struct config_llog_instance *cfg, + struct super_block *sb) { - struct config_llog_data *cld; - int rc; - ENTRY; + struct config_llog_data *cld; + int rc; - CDEBUG(D_MGC, "do adding config log %s:%p\n", logname, - cfg ? cfg->cfg_instance : 0); + ENTRY; - OBD_ALLOC(cld, sizeof(*cld) + strlen(logname) + 1); - if (!cld) - RETURN(ERR_PTR(-ENOMEM)); + CDEBUG(D_MGC, "do adding config log %s:%p\n", logname, + cfg ? cfg->cfg_instance : NULL); + + OBD_ALLOC(cld, sizeof(*cld) + strlen(logname) + 1); + if (!cld) + RETURN(ERR_PTR(-ENOMEM)); strcpy(cld->cld_logname, logname); if (cfg) @@ -223,32 +231,33 @@ struct config_llog_data *do_config_log_add(struct obd_device *obd, cld->cld_type = type; atomic_set(&cld->cld_refcount, 1); - /* Keep the mgc around until we are done */ - cld->cld_mgcexp = class_export_get(obd->obd_self_export); + /* Keep the mgc around until we are done */ + cld->cld_mgcexp = class_export_get(obd->obd_self_export); - if (cld_is_sptlrpc(cld)) { - sptlrpc_conf_log_start(logname); - cld->cld_cfg.cfg_obdname = obd->obd_name; - } + if (cld_is_sptlrpc(cld)) { + sptlrpc_conf_log_start(logname); + cld->cld_cfg.cfg_obdname = obd->obd_name; + } - rc = mgc_logname2resid(logname, &cld->cld_resid, type); + rc = mgc_logname2resid(logname, &cld->cld_resid, type); spin_lock(&config_list_lock); list_add(&cld->cld_list_chain, &config_llog_list); spin_unlock(&config_list_lock); - if (rc) { - config_log_put(cld); - RETURN(ERR_PTR(rc)); - } + if (rc) { + config_log_put(cld); + RETURN(ERR_PTR(rc)); + } - if (cld_is_sptlrpc(cld)) { - rc = mgc_process_log(obd, cld); + if (cld_is_sptlrpc(cld) || cld_is_nodemap(cld)) { + rc = mgc_process_log(obd, cld); if (rc && rc != -ENOENT) - CERROR("failed processing sptlrpc log: %d\n", rc); - } + CERROR("%s: failed processing log, type %d: rc = %d\n", + obd->obd_name, type, rc); + } - RETURN(cld); + RETURN(cld); } static struct config_llog_data *config_recover_log_add(struct obd_device *obd, @@ -305,62 +314,73 @@ static struct config_llog_data *config_params_log_add(struct obd_device *obd, * Each instance may be at a different point in the log. */ static int config_log_add(struct obd_device *obd, char *logname, - struct config_llog_instance *cfg, - struct super_block *sb) + struct config_llog_instance *cfg, + struct super_block *sb) { struct lustre_sb_info *lsi = s2lsi(sb); struct config_llog_data *cld; struct config_llog_data *sptlrpc_cld; struct config_llog_data *params_cld; + struct config_llog_data *nodemap_cld; char seclogname[32]; char *ptr; int rc; ENTRY; - CDEBUG(D_MGC, "adding config log %s:%p\n", logname, cfg->cfg_instance); + CDEBUG(D_MGC, "adding config log %s:%p\n", logname, cfg->cfg_instance); - /* - * for each regular log, the depended sptlrpc log name is - * -sptlrpc. multiple regular logs may share one sptlrpc log. - */ - ptr = strrchr(logname, '-'); - if (ptr == NULL || ptr - logname > 8) { - CERROR("logname %s is too long\n", logname); - RETURN(-EINVAL); - } + /* + * for each regular log, the depended sptlrpc log name is + * -sptlrpc. multiple regular logs may share one sptlrpc log. + */ + ptr = strrchr(logname, '-'); + if (ptr == NULL || ptr - logname > 8) { + CERROR("logname %s is too long\n", logname); + RETURN(-EINVAL); + } - memcpy(seclogname, logname, ptr - logname); - strcpy(seclogname + (ptr - logname), "-sptlrpc"); + memcpy(seclogname, logname, ptr - logname); + strcpy(seclogname + (ptr - logname), "-sptlrpc"); + + sptlrpc_cld = config_log_find(seclogname, NULL); + if (sptlrpc_cld == NULL) { + sptlrpc_cld = do_config_log_add(obd, seclogname, + CONFIG_T_SPTLRPC, NULL, NULL); + if (IS_ERR(sptlrpc_cld)) { + CERROR("can't create sptlrpc log: %s\n", seclogname); + GOTO(out, rc = PTR_ERR(sptlrpc_cld)); + } + } + + nodemap_cld = config_log_find(LUSTRE_NODEMAP_NAME, NULL); + if (!nodemap_cld && IS_SERVER(lsi) && !IS_MGS(lsi)) { + nodemap_cld = do_config_log_add(obd, LUSTRE_NODEMAP_NAME, + CONFIG_T_NODEMAP, NULL, NULL); + if (IS_ERR(nodemap_cld)) { + rc = PTR_ERR(nodemap_cld); + CERROR("%s: cannot create nodemap log: rc = %d\n", + obd->obd_name, rc); + GOTO(out_sptlrpc, rc); + } + } - sptlrpc_cld = config_log_find(seclogname, NULL); - if (sptlrpc_cld == NULL) { - sptlrpc_cld = do_config_log_add(obd, seclogname, - CONFIG_T_SPTLRPC, NULL, NULL); - if (IS_ERR(sptlrpc_cld)) { - CERROR("can't create sptlrpc log: %s\n", seclogname); - GOTO(out_err, rc = PTR_ERR(sptlrpc_cld)); - } - } params_cld = config_params_log_add(obd, cfg, sb); if (IS_ERR(params_cld)) { rc = PTR_ERR(params_cld); CERROR("%s: can't create params log: rc = %d\n", obd->obd_name, rc); - GOTO(out_err1, rc); + GOTO(out_nodemap, rc); } cld = do_config_log_add(obd, logname, CONFIG_T_CONFIG, cfg, sb); if (IS_ERR(cld)) { CERROR("can't create log: %s\n", logname); - GOTO(out_err2, rc = PTR_ERR(cld)); + GOTO(out_params, rc = PTR_ERR(cld)); } - cld->cld_sptlrpc = sptlrpc_cld; - cld->cld_params = params_cld; - - LASSERT(lsi->lsi_lmd); - if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)) { - struct config_llog_data *recover_cld; + LASSERT(lsi->lsi_lmd); + if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)) { + struct config_llog_data *recover_cld; ptr = strrchr(seclogname, '-'); if (ptr != NULL) { *ptr = 0; @@ -371,25 +391,32 @@ static int config_log_add(struct obd_device *obd, char *logname, config_log_put(cld); RETURN(-EINVAL); } - recover_cld = config_recover_log_add(obd, seclogname, cfg, sb); + recover_cld = config_recover_log_add(obd, seclogname, cfg, sb); if (IS_ERR(recover_cld)) - GOTO(out_err3, rc = PTR_ERR(recover_cld)); + GOTO(out_cld, rc = PTR_ERR(recover_cld)); cld->cld_recover = recover_cld; } + cld->cld_sptlrpc = sptlrpc_cld; + cld->cld_params = params_cld; + cld->cld_nodemap = nodemap_cld; + RETURN(0); -out_err3: +out_cld: config_log_put(cld); -out_err2: +out_params: config_log_put(params_cld); -out_err1: +out_nodemap: + config_log_put(nodemap_cld); + +out_sptlrpc: config_log_put(sptlrpc_cld); -out_err: - RETURN(rc); +out: + return rc; } DEFINE_MUTEX(llog_process_lock); @@ -398,36 +425,38 @@ DEFINE_MUTEX(llog_process_lock); */ static int config_log_end(char *logname, struct config_llog_instance *cfg) { - struct config_llog_data *cld; - struct config_llog_data *cld_sptlrpc = NULL; + struct config_llog_data *cld; + struct config_llog_data *cld_sptlrpc = NULL; struct config_llog_data *cld_params = NULL; - struct config_llog_data *cld_recover = NULL; - int rc = 0; - ENTRY; + struct config_llog_data *cld_recover = NULL; + struct config_llog_data *cld_nodemap = NULL; + int rc = 0; - cld = config_log_find(logname, cfg); - if (cld == NULL) - RETURN(-ENOENT); + ENTRY; + + cld = config_log_find(logname, cfg); + if (cld == NULL) + RETURN(-ENOENT); 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: - * the cld still hanging there waiting for lock cancel, and we - * remount again but failed in the middle and call log_end without - * calling start_log. - */ - if (unlikely(cld->cld_stopping)) { + /* + * 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: + * the cld still hanging there waiting for lock cancel, and we + * remount again but failed in the middle and call log_end without + * calling start_log. + */ + if (unlikely(cld->cld_stopping)) { mutex_unlock(&cld->cld_lock); - /* drop the ref from the find */ - config_log_put(cld); - RETURN(rc); - } + /* drop the ref from the find */ + config_log_put(cld); + RETURN(rc); + } - cld->cld_stopping = 1; + cld->cld_stopping = 1; - cld_recover = cld->cld_recover; - cld->cld_recover = NULL; + cld_recover = cld->cld_recover; + cld->cld_recover = NULL; mutex_unlock(&cld->cld_lock); if (cld_recover) { @@ -442,10 +471,12 @@ static int config_log_end(char *logname, struct config_llog_instance *cfg) cld->cld_sptlrpc = NULL; cld_params = cld->cld_params; cld->cld_params = NULL; + cld_nodemap = cld->cld_nodemap; + cld->cld_nodemap = NULL; spin_unlock(&config_list_lock); - if (cld_sptlrpc) - config_log_put(cld_sptlrpc); + if (cld_sptlrpc) + config_log_put(cld_sptlrpc); if (cld_params) { mutex_lock(&cld_params->cld_lock); @@ -454,17 +485,21 @@ static int config_log_end(char *logname, struct config_llog_instance *cfg) config_log_put(cld_params); } - /* drop the ref from the find */ - config_log_put(cld); - /* drop the start ref */ - config_log_put(cld); + /* don't set cld_stopping on nm lock as other targets may be active */ + if (cld_nodemap) + config_log_put(cld_nodemap); - CDEBUG(D_MGC, "end config log %s (%d)\n", logname ? logname : "client", - rc); - RETURN(rc); + /* drop the ref from the find */ + config_log_put(cld); + /* drop the start ref */ + config_log_put(cld); + + CDEBUG(D_MGC, "end config log %s (%d)\n", logname ? logname : "client", + rc); + RETURN(rc); } -#ifdef LPROCFS +#ifdef CONFIG_PROC_FS int lprocfs_mgc_rd_ir_state(struct seq_file *m, void *data) { struct obd_device *obd = data; @@ -487,8 +522,8 @@ int lprocfs_mgc_rd_ir_state(struct seq_file *m, void *data) if (cld->cld_recover == NULL) continue; seq_printf(m, " - { client: %s, nidtbl_version: %u }\n", - cld->cld_logname, - cld->cld_recover->cld_cfg.cfg_last_idx); + cld->cld_logname, + cld->cld_recover->cld_cfg.cfg_last_idx); } spin_unlock(&config_list_lock); @@ -515,10 +550,13 @@ static void do_requeue(struct config_llog_data *cld) LASSERT(atomic_read(&cld->cld_refcount) > 0); - /* Do not run mgc_process_log on a disconnected export or an + /* + * 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. */ - down_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem); + * semaphore to make the check non-racy. + */ + down_read_nested(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem, + OBD_CLI_SEM_MGC); if (cld->cld_mgcexp->exp_obd->u.cli.cl_conn_count != 0) { CDEBUG(D_MGC, "updating log %s\n", cld->cld_logname); rc = mgc_process_log(cld->cld_mgcexp->exp_obd, cld); @@ -708,13 +746,13 @@ static int mgc_local_llog_fini(const struct lu_env *env, RETURN(0); } -static int mgc_fs_setup(struct obd_device *obd, struct super_block *sb) +static int mgc_fs_setup(const struct lu_env *env, struct obd_device *obd, + struct super_block *sb) { struct lustre_sb_info *lsi = s2lsi(sb); struct client_obd *cli = &obd->u.cli; struct lu_fid rfid, fid; struct dt_object *root, *dto; - struct lu_env *env; int rc = 0; ENTRY; @@ -722,29 +760,21 @@ static int mgc_fs_setup(struct obd_device *obd, struct super_block *sb) LASSERT(lsi); LASSERT(lsi->lsi_dt_dev); - OBD_ALLOC_PTR(env); - if (env == NULL) - RETURN(-ENOMEM); - /* The mgc fs exclusion mutex. Only one fs can be setup at a time. */ mutex_lock(&cli->cl_mgc_mutex); /* Setup the configs dir */ - rc = lu_env_init(env, LCT_MG_THREAD); - if (rc) - GOTO(out_err, rc); - fid.f_seq = FID_SEQ_LOCAL_NAME; fid.f_oid = 1; fid.f_ver = 0; rc = local_oid_storage_init(env, lsi->lsi_dt_dev, &fid, &cli->cl_mgc_los); if (rc) - GOTO(out_env, rc); + RETURN(rc); rc = dt_root_get(env, lsi->lsi_dt_dev, &rfid); if (rc) - GOTO(out_env, rc); + GOTO(out_los, rc); root = dt_locate_at(env, lsi->lsi_dt_dev, &rfid, &cli->cl_mgc_los->los_dev->dd_lu_dev, NULL); @@ -782,37 +812,24 @@ out_los: cli->cl_mgc_los = NULL; mutex_unlock(&cli->cl_mgc_mutex); } -out_env: - lu_env_fini(env); -out_err: - OBD_FREE_PTR(env); return rc; } -static int mgc_fs_cleanup(struct obd_device *obd) +static int mgc_fs_cleanup(const struct lu_env *env, struct obd_device *obd) { - struct lu_env env; struct client_obd *cli = &obd->u.cli; - int rc; - ENTRY; LASSERT(cli->cl_mgc_los != NULL); - rc = lu_env_init(&env, LCT_MG_THREAD); - if (rc) - GOTO(unlock, rc); + mgc_local_llog_fini(env, obd); - mgc_local_llog_fini(&env, obd); - - lu_object_put_nocache(&env, &cli->cl_mgc_configs_dir->do_lu); + lu_object_put_nocache(env, &cli->cl_mgc_configs_dir->do_lu); cli->cl_mgc_configs_dir = NULL; - local_oid_storage_fini(&env, cli->cl_mgc_los); + local_oid_storage_fini(env, cli->cl_mgc_los); cli->cl_mgc_los = NULL; - lu_env_fini(&env); -unlock: class_decref(obd, "mgc_fs", obd); mutex_unlock(&cli->cl_mgc_mutex); @@ -857,36 +874,34 @@ static int mgc_llog_fini(const struct lu_env *env, struct obd_device *obd) static atomic_t mgc_count = ATOMIC_INIT(0); -static int mgc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) +static int mgc_precleanup(struct obd_device *obd) { - int rc = 0; - int temp; + int rc = 0; + int temp; ENTRY; - switch (stage) { - case OBD_CLEANUP_EARLY: - break; - case OBD_CLEANUP_EXPORTS: - if (atomic_dec_and_test(&mgc_count)) { - LASSERT(rq_state & RQ_RUNNING); - /* stop requeue thread */ - temp = RQ_STOP; - } else { - /* wakeup requeue thread to clean our cld */ - temp = RQ_NOW | RQ_PRECLEANUP; - } - spin_lock(&config_list_lock); - rq_state |= temp; - spin_unlock(&config_list_lock); - wake_up(&rq_waitq); - if (temp & RQ_STOP) - wait_for_completion(&rq_exit); - obd_cleanup_client_import(obd); - rc = mgc_llog_fini(NULL, obd); - if (rc != 0) - CERROR("failed to cleanup llogging subsystems\n"); - break; + if (atomic_dec_and_test(&mgc_count)) { + LASSERT(rq_state & RQ_RUNNING); + /* stop requeue thread */ + temp = RQ_STOP; + } else { + /* wakeup requeue thread to clean our cld */ + temp = RQ_NOW | RQ_PRECLEANUP; } + + spin_lock(&config_list_lock); + rq_state |= temp; + spin_unlock(&config_list_lock); + wake_up(&rq_waitq); + + if (temp & RQ_STOP) + wait_for_completion(&rq_exit); + obd_cleanup_client_import(obd); + + rc = mgc_llog_fini(NULL, obd); + if (rc != 0) + CERROR("failed to cleanup llogging subsystems\n"); + RETURN(rc); } @@ -928,7 +943,7 @@ static int mgc_setup(struct obd_device *obd, struct lustre_cfg *lcfg) GOTO(err_cleanup, rc); } -#ifdef LPROCFS +#ifdef CONFIG_PROC_FS obd->obd_vars = lprocfs_mgc_obd_vars; lprocfs_obd_setup(obd); #endif @@ -1001,8 +1016,8 @@ static int mgc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, } /* Make sure not to re-enqueue when the mgc is stopping (we get called from client_disconnect_export) */ - if (!lock->l_conn_export || - !lock->l_conn_export->exp_obd->u.cli.cl_conn_count) { + if (lock->l_conn_export == NULL || + lock->l_conn_export->exp_obd->u.cli.cl_conn_count == 0) { CDEBUG(D_MGC, "log %.8s: disconnecting, won't requeue\n", cld->cld_logname); config_log_put(cld); @@ -1066,8 +1081,8 @@ 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, +static int mgc_enqueue(struct obd_export *exp, enum ldlm_type type, + union ldlm_policy_data *policy, enum ldlm_mode mode, __u64 *flags, void *bl_cb, void *cp_cb, void *gl_cb, void *data, __u32 lvb_len, void *lvb_swabber, struct lustre_handle *lockh) @@ -1114,14 +1129,14 @@ static int mgc_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm, RETURN(rc); } -static int mgc_cancel(struct obd_export *exp, ldlm_mode_t mode, +static int mgc_cancel(struct obd_export *exp, enum ldlm_mode mode, struct lustre_handle *lockh) { - ENTRY; + ENTRY; - ldlm_lock_decref(lockh, mode); + ldlm_lock_decref(lockh, mode); - RETURN(0); + RETURN(0); } static void mgc_notify_active(struct obd_device *unused) @@ -1218,13 +1233,13 @@ static int mgc_set_info_async(const struct lu_env *env, struct obd_export *exp, if (vallen != sizeof(struct super_block)) RETURN(-EINVAL); - rc = mgc_fs_setup(exp->exp_obd, sb); + rc = mgc_fs_setup(env, exp->exp_obd, sb); RETURN(rc); } if (KEY_IS(KEY_CLEAR_FS)) { if (vallen != 0) RETURN(-EINVAL); - rc = mgc_fs_cleanup(exp->exp_obd); + rc = mgc_fs_cleanup(env, exp->exp_obd); RETURN(rc); } if (KEY_IS(KEY_SET_INFO)) { @@ -1282,9 +1297,7 @@ static int mgc_set_info_async(const struct lu_env *env, struct obd_export *exp, } static int mgc_get_info(const struct lu_env *env, struct obd_export *exp, - __u32 keylen, void *key, - __u32 *vallen, void *val, - struct lov_stripe_md *unused) + __u32 keylen, void *key, __u32 *vallen, void *val) { int rc = -EINVAL; @@ -1554,24 +1567,29 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, /** * This function is called if this client was notified for target restarting - * by the MGS. A CONFIG_READ RPC is going to send to fetch recovery logs. + * by the MGS. A CONFIG_READ RPC is going to send to fetch recovery or + * nodemap logs. */ -static int mgc_process_recover_log(struct obd_device *obd, - struct config_llog_data *cld) +static int mgc_process_recover_nodemap_log(struct obd_device *obd, + struct config_llog_data *cld) { - struct ptlrpc_request *req = NULL; - struct config_llog_instance *cfg = &cld->cld_cfg; - struct mgs_config_body *body; - struct mgs_config_res *res; - struct ptlrpc_bulk_desc *desc; + struct ptlrpc_request *req = NULL; + struct config_llog_instance *cfg = &cld->cld_cfg; + struct mgs_config_body *body; + struct mgs_config_res *res; + struct nodemap_config *new_config = NULL; + struct lu_nodemap *recent_nodemap = NULL; + struct ptlrpc_bulk_desc *desc; struct page **pages; - int nrpages; - bool eof = true; + __u64 config_read_offset = 0; + int nrpages; + bool eof = true; bool mne_swab = false; - int i; - int ealen; - int rc; - ENTRY; + int i; + int ealen; + int rc; + + ENTRY; /* allocate buffer for bulk transfer. * if this is the first time for this mgs to read logs, @@ -1580,7 +1598,7 @@ static int mgc_process_recover_log(struct obd_device *obd, * small and CONFIG_READ_NRPAGES will be used. */ nrpages = CONFIG_READ_NRPAGES; - if (cfg->cfg_last_idx == 0) /* the first time */ + if (cfg->cfg_last_idx == 0 || cld_is_nodemap(cld)) nrpages = CONFIG_READ_NRPAGES_INIT; OBD_ALLOC(pages, sizeof(*pages) * nrpages); @@ -1588,69 +1606,108 @@ static int mgc_process_recover_log(struct obd_device *obd, GOTO(out, rc = -ENOMEM); for (i = 0; i < nrpages; i++) { - pages[i] = alloc_page(GFP_IOFS); + pages[i] = alloc_page(GFP_KERNEL); if (pages[i] == NULL) GOTO(out, rc = -ENOMEM); } again: - LASSERT(cld_is_recover(cld)); +#ifdef HAVE_SERVER_SUPPORT + if (cld_is_nodemap(cld) && config_read_offset == 0) { + new_config = nodemap_config_alloc(); + if (IS_ERR(new_config)) { + rc = PTR_ERR(new_config); + new_config = NULL; + GOTO(out, rc); + } + } +#endif + LASSERT(cld_is_recover(cld) || cld_is_nodemap(cld)); LASSERT(mutex_is_locked(&cld->cld_lock)); - req = ptlrpc_request_alloc(class_exp2cliimp(cld->cld_mgcexp), - &RQF_MGS_CONFIG_READ); - if (req == NULL) - GOTO(out, rc = -ENOMEM); + req = ptlrpc_request_alloc(class_exp2cliimp(cld->cld_mgcexp), + &RQF_MGS_CONFIG_READ); + if (req == NULL) + GOTO(out, rc = -ENOMEM); - rc = ptlrpc_request_pack(req, LUSTRE_MGS_VERSION, MGS_CONFIG_READ); - if (rc) - GOTO(out, rc); + rc = ptlrpc_request_pack(req, LUSTRE_MGS_VERSION, MGS_CONFIG_READ); + if (rc) + GOTO(out, rc); - /* pack request */ - body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY); - LASSERT(body != NULL); - LASSERT(sizeof(body->mcb_name) > strlen(cld->cld_logname)); + /* pack request */ + body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY); + LASSERT(body != NULL); + LASSERT(sizeof(body->mcb_name) > strlen(cld->cld_logname)); if (strlcpy(body->mcb_name, cld->cld_logname, sizeof(body->mcb_name)) >= sizeof(body->mcb_name)) GOTO(out, rc = -E2BIG); - body->mcb_offset = cfg->cfg_last_idx + 1; - body->mcb_type = cld->cld_type; + if (cld_is_nodemap(cld)) + body->mcb_offset = config_read_offset; + else + body->mcb_offset = cfg->cfg_last_idx + 1; + body->mcb_type = cld->cld_type; body->mcb_bits = PAGE_CACHE_SHIFT; - body->mcb_units = nrpages; + body->mcb_units = nrpages; /* allocate bulk transfer descriptor */ - desc = ptlrpc_prep_bulk_imp(req, nrpages, 1, BULK_PUT_SINK, - MGS_BULK_PORTAL); + desc = ptlrpc_prep_bulk_imp(req, nrpages, 1, + PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV, + MGS_BULK_PORTAL, + &ptlrpc_bulk_kiov_pin_ops); if (desc == NULL) GOTO(out, rc = -ENOMEM); for (i = 0; i < nrpages; i++) - ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE); + desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0, + PAGE_CACHE_SIZE); - ptlrpc_request_set_replen(req); - rc = ptlrpc_queue_wait(req); - if (rc) - GOTO(out, rc); + ptlrpc_request_set_replen(req); + rc = ptlrpc_queue_wait(req); + if (rc) + GOTO(out, rc); - res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES); - if (res->mcr_size < res->mcr_offset) - GOTO(out, rc = -EINVAL); + res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES); + if (!res) + GOTO(out, rc = -EPROTO); - /* always update the index even though it might have errors with - * handling the recover logs */ - cfg->cfg_last_idx = res->mcr_offset; - eof = res->mcr_offset == res->mcr_size; + if (cld_is_nodemap(cld)) { + config_read_offset = res->mcr_offset; + eof = config_read_offset == II_END_OFF; + } else { + if (res->mcr_size < res->mcr_offset) + GOTO(out, rc = -EINVAL); + + /* always update the index even though it might have errors with + * handling the recover logs + */ + cfg->cfg_last_idx = res->mcr_offset; + eof = res->mcr_offset == res->mcr_size; - CDEBUG(D_INFO, "Latest version "LPD64", more %d.\n", - res->mcr_offset, eof == false); + CDEBUG(D_INFO, "Latest version "LPD64", more %d.\n", + res->mcr_offset, eof == false); + } - ealen = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, 0); - if (ealen < 0) - GOTO(out, rc = ealen); + ealen = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, 0); + if (ealen < 0) + GOTO(out, rc = ealen); if (ealen > nrpages << PAGE_CACHE_SHIFT) GOTO(out, rc = -EINVAL); if (ealen == 0) { /* no logs transferred */ +#ifdef HAVE_SERVER_SUPPORT + /* config changed since first read RPC */ + if (cld_is_nodemap(cld) && config_read_offset == 0) { + recent_nodemap = NULL; + nodemap_config_dealloc(new_config); + new_config = NULL; + + CDEBUG(D_INFO, "nodemap config changed in transit, retrying\n"); + + /* setting eof to false, we request config again */ + eof = false; + GOTO(out, rc = 0); + } +#endif if (!eof) rc = -EINVAL; GOTO(out, rc); @@ -1664,30 +1721,60 @@ again: mne_swab = !mne_swab; #endif + /* When a nodemap config is received, we build a new nodemap config, + * with new nodemap structs. We keep track of the most recently added + * nodemap since the config is read ordered by nodemap_id, and so it + * is likely that the next record will be related. Because access to + * the nodemaps is single threaded until the nodemap_config is active, + * we don't need to reference count with recent_nodemap, though + * recent_nodemap should be set to NULL when the nodemap_config + * is either destroyed or set active. + */ for (i = 0; i < nrpages && ealen > 0; i++) { int rc2; - void *ptr; + union lu_page *ptr; ptr = kmap(pages[i]); - rc2 = mgc_apply_recover_logs(obd, cld, res->mcr_offset, ptr, - min_t(int, ealen, PAGE_CACHE_SIZE), - mne_swab); + if (cld_is_nodemap(cld)) + rc2 = nodemap_process_idx_pages(new_config, ptr, + &recent_nodemap); + else + rc2 = mgc_apply_recover_logs(obd, cld, res->mcr_offset, + ptr, + min_t(int, ealen, + PAGE_CACHE_SIZE), + mne_swab); kunmap(pages[i]); if (rc2 < 0) { - CWARN("Process recover log %s error %d\n", - cld->cld_logname, rc2); + CWARN("%s: error processing %s log %s: rc = %d\n", + obd->obd_name, + cld_is_nodemap(cld) ? "nodemap" : "recovery", + cld->cld_logname, + rc2); break; - } + } ealen -= PAGE_CACHE_SIZE; - } + } out: - if (req) - ptlrpc_req_finished(req); + if (req) { + ptlrpc_req_finished(req); + req = NULL; + } - if (rc == 0 && !eof) - goto again; + if (rc == 0 && !eof) + goto again; + +#ifdef HAVE_SERVER_SUPPORT + if (new_config != NULL) { + /* recent_nodemap cannot be used after set_active/dealloc */ + if (rc == 0) + nodemap_config_set_active_mgc(new_config); + else + nodemap_config_dealloc(new_config); + } +#endif if (pages) { for (i = 0; i < nrpages; i++) { @@ -1717,7 +1804,7 @@ static int mgc_llog_local_copy(const struct lu_env *env, * - if failed then move bakup to logname again */ - OBD_ALLOC(temp_log, strlen(logname) + 1); + OBD_ALLOC(temp_log, strlen(logname) + 2); if (!temp_log) RETURN(-ENOMEM); sprintf(temp_log, "%sT", logname); @@ -1739,7 +1826,7 @@ out: obd->obd_name, logname, rc); } llog_erase(env, lctxt, NULL, temp_log); - OBD_FREE(temp_log, strlen(logname) + 1); + OBD_FREE(temp_log, strlen(logname) + 2); return rc; } @@ -1853,23 +1940,62 @@ out_free: return rc; } -/** Get a config log from the MGS and process it. - * This func is called for both clients and servers. - * Copy the log locally before parsing it if appropriate (non-MGS server) +static bool mgc_import_in_recovery(struct obd_import *imp) +{ + bool in_recovery = true; + + spin_lock(&imp->imp_lock); + if (imp->imp_state == LUSTRE_IMP_FULL || + imp->imp_state == LUSTRE_IMP_CLOSED) + in_recovery = false; + spin_unlock(&imp->imp_lock); + + return in_recovery; +} + +/** + * Get a configuration log from the MGS and process it. + * + * This function is called for both clients and servers to process the + * configuration log from the MGS. The MGC enqueues a DLM lock on the + * log from the MGS, and if the lock gets revoked the MGC will be notified + * by the lock cancellation callback that the config log has changed, + * and will enqueue another MGS lock on it, and then continue processing + * the new additions to the end of the log. + * + * Since the MGC import is not replayable, if the import is being evicted + * (rcl == -ESHUTDOWN, \see ptlrpc_import_delay_req()), retry to process + * the log until recovery is finished or the import is closed. + * + * Make a local copy of the log before parsing it if appropriate (non-MGS + * server) so that the server can start even when the MGS is down. + * + * There shouldn't be multiple processes running process_log at once -- + * sounds like badness. It actually might be fine, as long as they're not + * trying to update from the same log simultaneously, in which case we + * should use a per-log semaphore instead of cld_lock. + * + * \param[in] mgc MGC device by which to fetch the configuration log + * \param[in] cld log processing state (stored in lock callback data) + * + * \retval 0 on success + * \retval negative errno on failure */ int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld) { struct lustre_handle lockh = { 0 }; __u64 flags = LDLM_FL_NO_LRU; int rc = 0, rcl; + bool retry = false; ENTRY; - LASSERT(cld); + LASSERT(cld != NULL); /* I don't want multiple processes running process_log at once -- 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.) */ +restart: mutex_lock(&cld->cld_lock); if (cld->cld_stopping) { mutex_unlock(&cld->cld_lock); @@ -1878,39 +2004,79 @@ int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld) OBD_FAIL_TIMEOUT(OBD_FAIL_MGC_PAUSE_PROCESS_LOG, 20); - CDEBUG(D_MGC, "Process log %s:%p from %d\n", cld->cld_logname, - cld->cld_cfg.cfg_instance, cld->cld_cfg.cfg_last_idx + 1); - - /* Get the cfg lock on the llog */ - rcl = mgc_enqueue(mgc->u.cli.cl_mgc_mgsexp, NULL, LDLM_PLAIN, NULL, - LCK_CR, &flags, NULL, NULL, NULL, - cld, 0, NULL, &lockh); - if (rcl == 0) { - /* Get the cld, it will be released in mgc_blocking_ast. */ - config_log_get(cld); - rc = ldlm_lock_set_data(&lockh, (void *)cld); - LASSERT(rc == 0); - } else { - CDEBUG(D_MGC, "Can't get cfg lock: %d\n", rcl); + CDEBUG(D_MGC, "Process log %s:%p from %d\n", cld->cld_logname, + cld->cld_cfg.cfg_instance, cld->cld_cfg.cfg_last_idx + 1); + + /* Get the cfg lock on the llog */ + rcl = mgc_enqueue(mgc->u.cli.cl_mgc_mgsexp, LDLM_PLAIN, NULL, + LCK_CR, &flags, NULL, NULL, NULL, + cld, 0, NULL, &lockh); + if (rcl == 0) { + /* Get the cld, it will be released in mgc_blocking_ast. */ + config_log_get(cld); + rc = ldlm_lock_set_data(&lockh, (void *)cld); + LASSERT(rc == 0); + } else { + CDEBUG(D_MGC, "Can't get cfg lock: %d\n", rcl); + + if (rcl == -ESHUTDOWN && + atomic_read(&mgc->u.cli.cl_mgc_refcount) > 0 && !retry) { + struct obd_import *imp; + struct l_wait_info lwi; + int secs = cfs_time_seconds(obd_timeout); + + mutex_unlock(&cld->cld_lock); + imp = class_exp2cliimp(mgc->u.cli.cl_mgc_mgsexp); + + /* Let's force the pinger, and wait the import to be + * connected, note: since mgc import is non-replayable, + * and even the import state is disconnected, it does + * not mean the "recovery" is stopped, so we will keep + * waitting until timeout or the import state is + * FULL or closed */ + ptlrpc_pinger_force(imp); + + lwi = LWI_TIMEOUT(secs, NULL, NULL); + l_wait_event(imp->imp_recovery_waitq, + !mgc_import_in_recovery(imp), &lwi); + + if (imp->imp_state == LUSTRE_IMP_FULL) { + retry = true; + goto restart; + } else { + mutex_lock(&cld->cld_lock); + cld->cld_lostlock = 1; + } + } else { + /* mark cld_lostlock so that it will requeue + * after MGC becomes available. */ + cld->cld_lostlock = 1; + } + /* Get extra reference, it will be put in requeue thread */ + config_log_get(cld); + } - /* mark cld_lostlock so that it will requeue - * after MGC becomes available. */ - cld->cld_lostlock = 1; - /* Get extra reference, it will be put in requeue thread */ - config_log_get(cld); - } + if (cld_is_recover(cld) || cld_is_nodemap(cld)) { + if (!rcl) + rc = mgc_process_recover_nodemap_log(mgc, cld); + else if (cld_is_nodemap(cld)) + rc = rcl; - if (cld_is_recover(cld)) { - rc = 0; /* this is not a fatal error for recover log */ - if (rcl == 0) - rc = mgc_process_recover_log(mgc, cld); - } else { - rc = mgc_process_cfg_log(mgc, cld, rcl != 0); - } + if (cld_is_recover(cld) && rc) { + if (!rcl) { + CERROR("%s: recover log %s failed, not fatal: rc = %d\n", + mgc->obd_name, cld->cld_logname, rc); + cld->cld_lostlock = 1; + } + rc = 0; /* this is not a fatal error for recover log */ + } + } else { + rc = mgc_process_cfg_log(mgc, cld, rcl != 0); + } - CDEBUG(D_MGC, "%s: configuration from log '%s' %sed (%d).\n", - mgc->obd_name, cld->cld_logname, rc ? "fail" : "succeed", rc); + CDEBUG(D_MGC, "%s: configuration from log '%s' %sed (%d).\n", + mgc->obd_name, cld->cld_logname, rc ? "fail" : "succeed", rc); mutex_unlock(&cld->cld_lock); @@ -2057,13 +2223,14 @@ static int __init mgc_init(void) LUSTRE_MGC_NAME, NULL); } -static void /*__exit*/ mgc_exit(void) +static void __exit mgc_exit(void) { class_unregister_type(LUSTRE_MGC_NAME); } -MODULE_AUTHOR("Sun Microsystems, Inc. "); +MODULE_AUTHOR("OpenSFS, Inc. "); MODULE_DESCRIPTION("Lustre Management Client"); +MODULE_VERSION(LUSTRE_VERSION_STRING); MODULE_LICENSE("GPL"); module_init(mgc_init);