X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmgc%2Fmgc_request.c;h=b263c4b4cce1e83de0867a1e98e4df1ecd905f24;hp=8b208e7b3369c1788b25805eee4a82a235a5bcac;hb=45ce1044cd7b94621e1161cd23c600f8e1c18317;hpb=1d2bf12a7c3e954858e08134f4e7bc1fce9a5812 diff --git a/lustre/mgc/mgc_request.c b/lustre/mgc/mgc_request.c index 8b208e7..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, Whamcloud, Inc. + * Copyright (c) 2011, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -41,21 +41,18 @@ #define DEBUG_SUBSYSTEM S_MGC #define D_MGC D_CONFIG /*|D_WARNING*/ -#ifdef __KERNEL__ -# include -# include -# include -# include -#else -# include -#endif +#include +#include -#include -#include +#include #include -#include -#include +#include #include +#include +#include +#include +#include + #include "mgc_internal.h" static int mgc_name2resid(char *name, int len, struct ldlm_res_id *res_id, @@ -63,7 +60,7 @@ static int mgc_name2resid(char *name, int len, struct ldlm_res_id *res_id, { __u64 resname = 0; - if (len > 8) { + if (len > sizeof(resname)) { CERROR("name too long: %s\n", name); return -EINVAL; } @@ -82,9 +79,11 @@ static int mgc_name2resid(char *name, int len, struct ldlm_res_id *res_id, case CONFIG_T_SPTLRPC: resname = 0; break; - case CONFIG_T_RECOVER: - resname = type; - break; + case CONFIG_T_RECOVER: + case CONFIG_T_PARAMS: + case CONFIG_T_NODEMAP: + resname = type; + break; default: LBUG(); } @@ -102,62 +101,70 @@ int mgc_fsname2resid(char *fsname, struct ldlm_res_id *res_id, int type) } EXPORT_SYMBOL(mgc_fsname2resid); -int mgc_logname2resid(char *logname, struct ldlm_res_id *res_id, int type) +static int mgc_logname2resid(char *logname, struct ldlm_res_id *res_id, + int type) { - char *name_end; - int len; - - /* logname consists of "fsname-nodetype". - * e.g. "lustre-MDT0001", "SUN-000-client" */ - name_end = strrchr(logname, '-'); - LASSERT(name_end); - len = name_end - logname; - return mgc_name2resid(logname, len, res_id, type); + char *name_end; + int len; + + /* logname consists of "fsname-nodetype". + * e.g. "lustre-MDT0001", "SUN-000-client" + * there is an exception: llog "params" */ + name_end = strrchr(logname, '-'); + if (!name_end) + len = strlen(logname); + else + len = name_end - logname; + return mgc_name2resid(logname, len, res_id, type); } /********************** config llog list **********************/ -static CFS_LIST_HEAD(config_llog_list); +static struct list_head config_llog_list = LIST_HEAD_INIT(config_llog_list); static DEFINE_SPINLOCK(config_list_lock); /* Take a reference to a config log */ static int config_log_get(struct config_llog_data *cld) { - ENTRY; - cfs_atomic_inc(&cld->cld_refcount); - CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname, - cfs_atomic_read(&cld->cld_refcount)); - RETURN(0); + ENTRY; + atomic_inc(&cld->cld_refcount); + CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname, + atomic_read(&cld->cld_refcount)); + RETURN(0); } /* Drop a reference to a config log. When no longer referenced, we can free the config log data */ static void config_log_put(struct config_llog_data *cld) { - ENTRY; - - CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname, - cfs_atomic_read(&cld->cld_refcount)); - LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0); - - /* 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); - - 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_is_sptlrpc(cld)) - sptlrpc_conf_log_stop(cld->cld_logname); + ENTRY; - class_export_put(cld->cld_mgcexp); - OBD_FREE(cld, sizeof(*cld) + strlen(cld->cld_logname) + 1); - } + CDEBUG(D_INFO, "log %s refs %d\n", cld->cld_logname, + atomic_read(&cld->cld_refcount)); + LASSERT(atomic_read(&cld->cld_refcount) > 0); + + /* spinlock to make sure no item with 0 refcount in the list */ + if (atomic_dec_and_lock(&cld->cld_refcount, &config_list_lock)) { + list_del(&cld->cld_list_chain); + spin_unlock(&config_list_lock); + + 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_params) + config_log_put(cld->cld_params); + 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); + } - EXIT; + EXIT; } /* Find a config log by name */ @@ -173,80 +180,84 @@ struct config_llog_data *config_log_find(char *logname, LASSERT(logname != NULL); instance = cfg ? cfg->cfg_instance : NULL; - cfs_spin_lock(&config_list_lock); - cfs_list_for_each_entry(cld, &config_llog_list, cld_list_chain) { + spin_lock(&config_list_lock); + list_for_each_entry(cld, &config_llog_list, cld_list_chain) { /* check if instance equals */ if (instance != cld->cld_cfg.cfg_instance) continue; - /* instance may be NULL, should check name */ - if (strcmp(logname, cld->cld_logname) == 0) { - found = cld; - break; - } - } - if (found) { - cfs_atomic_inc(&found->cld_refcount); - LASSERT(found->cld_stopping == 0 || cld_is_sptlrpc(found) == 0); - } - cfs_spin_unlock(&config_list_lock); - RETURN(found); + /* instance may be NULL, should check name */ + if (strcmp(logname, cld->cld_logname) == 0) { + found = cld; + break; + } + } + if (found) { + atomic_inc(&found->cld_refcount); + LASSERT(found->cld_stopping == 0 || cld_is_sptlrpc(found) == 0); + } + spin_unlock(&config_list_lock); + RETURN(found); } 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); - - OBD_ALLOC(cld, sizeof(*cld) + strlen(logname) + 1); - if (!cld) - RETURN(ERR_PTR(-ENOMEM)); - - strcpy(cld->cld_logname, logname); - if (cfg) - cld->cld_cfg = *cfg; - cfs_mutex_init(&cld->cld_lock); - cld->cld_cfg.cfg_last_idx = 0; - cld->cld_cfg.cfg_flags = 0; - cld->cld_cfg.cfg_sb = sb; - cld->cld_type = type; - cfs_atomic_set(&cld->cld_refcount, 1); - - /* 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; - } + ENTRY; - rc = mgc_logname2resid(logname, &cld->cld_resid, type); + 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) + cld->cld_cfg = *cfg; + 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; + 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); + + if (cld_is_sptlrpc(cld)) { + sptlrpc_conf_log_start(logname); + cld->cld_cfg.cfg_obdname = obd->obd_name; + } - cfs_spin_lock(&config_list_lock); - cfs_list_add(&cld->cld_list_chain, &config_llog_list); - cfs_spin_unlock(&config_list_lock); + rc = mgc_logname2resid(logname, &cld->cld_resid, type); - if (rc) { - config_log_put(cld); - RETURN(ERR_PTR(rc)); - } + spin_lock(&config_list_lock); + list_add(&cld->cld_list_chain, &config_llog_list); + spin_unlock(&config_list_lock); - if (cld_is_sptlrpc(cld)) { - rc = mgc_process_log(obd, cld); - if (rc) - CERROR("failed processing sptlrpc log: %d\n", rc); - } + if (rc) { + config_log_put(cld); + RETURN(ERR_PTR(rc)); + } - RETURN(cld); + if (cld_is_sptlrpc(cld) || cld_is_nodemap(cld)) { + rc = mgc_process_log(obd, cld); + if (rc && rc != -ENOENT) + CERROR("%s: failed processing log, type %d: rc = %d\n", + obd->obd_name, type, rc); + } + + RETURN(cld); } static struct config_llog_data *config_recover_log_add(struct obd_device *obd, @@ -262,6 +273,10 @@ static struct config_llog_data *config_recover_log_add(struct obd_device *obd, 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); @@ -279,6 +294,19 @@ static struct config_llog_data *config_recover_log_add(struct obd_device *obd, return cld; } +static struct config_llog_data *config_params_log_add(struct obd_device *obd, + struct config_llog_instance *cfg, struct super_block *sb) +{ + struct config_llog_instance lcfg = *cfg; + struct config_llog_data *cld; + + lcfg.cfg_instance = sb; + + cld = do_config_log_add(obd, PARAMS_FILENAME, CONFIG_T_PARAMS, + &lcfg, sb); + + return cld; +} /** Add this log to the list of active logs watched by an MGC. * Active means we're watching for updates. @@ -286,181 +314,259 @@ static struct config_llog_data *config_recover_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; - char seclogname[32]; - char *ptr; - ENTRY; + 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); - RETURN(PTR_ERR(sptlrpc_cld)); - } - } + 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)); + } + } - cld = do_config_log_add(obd, logname, CONFIG_T_CONFIG, cfg, sb); - if (IS_ERR(cld)) { - CERROR("can't create log: %s\n", logname); - config_log_put(sptlrpc_cld); - RETURN(PTR_ERR(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); + } + } - cld->cld_sptlrpc = 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_nodemap, rc); + } - LASSERT(lsi->lsi_lmd); - if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)) { - struct config_llog_data *recover_cld; - *strrchr(seclogname, '-') = 0; - recover_cld = config_recover_log_add(obd, seclogname, cfg, sb); - if (IS_ERR(recover_cld)) { - config_log_put(cld); - RETURN(PTR_ERR(recover_cld)); - } - cld->cld_recover = recover_cld; - } + 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_params, rc = PTR_ERR(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; + } + else { + CERROR("%s: sptlrpc log name not correct, %s: " + "rc = %d\n", obd->obd_name, seclogname, -EINVAL); + config_log_put(cld); + RETURN(-EINVAL); + } + recover_cld = config_recover_log_add(obd, seclogname, cfg, sb); + if (IS_ERR(recover_cld)) + GOTO(out_cld, rc = PTR_ERR(recover_cld)); + cld->cld_recover = recover_cld; + } - RETURN(0); + cld->cld_sptlrpc = sptlrpc_cld; + cld->cld_params = params_cld; + cld->cld_nodemap = nodemap_cld; + + RETURN(0); + +out_cld: + config_log_put(cld); + +out_params: + config_log_put(params_cld); + +out_nodemap: + config_log_put(nodemap_cld); + +out_sptlrpc: + config_log_put(sptlrpc_cld); + +out: + return rc; } -CFS_DEFINE_MUTEX(llog_process_lock); +DEFINE_MUTEX(llog_process_lock); /** Stop watching for updates on this log. */ 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_recover = NULL; - int rc = 0; - ENTRY; + 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; + struct config_llog_data *cld_nodemap = NULL; + int rc = 0; - cld = config_log_find(logname, cfg); - if (cld == NULL) - RETURN(-ENOENT); - - cfs_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)) { - cfs_mutex_unlock(&cld->cld_lock); - /* drop the ref from the find */ - config_log_put(cld); - RETURN(rc); - } + ENTRY; - cld->cld_stopping = 1; + 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)) { + mutex_unlock(&cld->cld_lock); + /* drop the ref from the find */ + config_log_put(cld); + RETURN(rc); + } - cld_recover = cld->cld_recover; - cld->cld_recover = NULL; - cfs_mutex_unlock(&cld->cld_lock); + cld->cld_stopping = 1; - 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); - } + cld_recover = cld->cld_recover; + cld->cld_recover = NULL; + mutex_unlock(&cld->cld_lock); + + 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; + 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_params) { + mutex_lock(&cld_params->cld_lock); + cld_params->cld_stopping = 1; + mutex_unlock(&cld_params->cld_lock); + config_log_put(cld_params); + } - if (cld_sptlrpc) - config_log_put(cld_sptlrpc); + /* don't set cld_stopping on nm lock as other targets may be active */ + if (cld_nodemap) + config_log_put(cld_nodemap); - /* drop the ref from the find */ - config_log_put(cld); - /* drop the start ref */ - config_log_put(cld); + /* 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); + CDEBUG(D_MGC, "end config log %s (%d)\n", logname ? logname : "client", + rc); + RETURN(rc); } -int lprocfs_mgc_rd_ir_state(char *page, char **start, off_t off, - int count, int *eof, void *data) +#ifdef CONFIG_PROC_FS +int lprocfs_mgc_rd_ir_state(struct seq_file *m, void *data) { - struct obd_device *obd = data; - struct obd_import *imp = obd->u.cli.cl_import; - struct obd_connect_data *ocd = &imp->imp_connect_data; - struct config_llog_data *cld; - int rc = 0; - ENTRY; - - rc = snprintf(page, count, "imperative_recovery: %s\n", - OCD_HAS_FLAG(ocd, IMP_RECOV) ? "ENABLED" : "DISABLED"); - rc += snprintf(page + rc, count - rc, "client_state:\n"); + struct obd_device *obd = data; + struct obd_import *imp; + struct obd_connect_data *ocd; + struct config_llog_data *cld; + ENTRY; - 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); + LASSERT(obd != NULL); + LPROCFS_CLIMP_CHECK(obd); + imp = obd->u.cli.cl_import; + ocd = &imp->imp_connect_data; + + seq_printf(m, "imperative_recovery: %s\n", + OCD_HAS_FLAG(ocd, IMP_RECOV) ? "ENABLED" : "DISABLED"); + seq_printf(m, "client_state:\n"); + + spin_lock(&config_list_lock); + list_for_each_entry(cld, &config_llog_list, cld_list_chain) { + 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); + } + spin_unlock(&config_list_lock); - RETURN(rc); + LPROCFS_CLIMP_EXIT(obd); + RETURN(0); } +#endif /* reenqueue any lost locks */ -#define RQ_RUNNING 0x1 -#define RQ_NOW 0x2 -#define RQ_LATER 0x4 -#define RQ_STOP 0x8 +#define RQ_RUNNING 0x1 +#define RQ_NOW 0x2 +#define RQ_LATER 0x4 +#define RQ_STOP 0x8 +#define RQ_PRECLEANUP 0x10 static int rq_state = 0; -static cfs_waitq_t rq_waitq; -static CFS_DECLARE_COMPLETION(rq_exit); +static wait_queue_head_t rq_waitq; +static DECLARE_COMPLETION(rq_exit); +static DECLARE_COMPLETION(rq_start); static void do_requeue(struct config_llog_data *cld) { - ENTRY; - LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0); - - /* 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); - 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); - } else { - 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); + int rc = 0; + ENTRY; + + LASSERT(atomic_read(&cld->cld_refcount) > 0); + + /* + * 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_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); + if (rc && rc != -ENOENT) + CERROR("failed processing log: %d\n", rc); + } else { + CDEBUG(D_MGC, "disconnecting, won't update log %s\n", + cld->cld_logname); + } + up_read(&cld->cld_mgcexp->exp_obd->u.cli.cl_sem); EXIT; } @@ -474,35 +580,40 @@ static void do_requeue(struct config_llog_data *cld) static int mgc_requeue_thread(void *data) { - char name[] = "ll_cfg_requeue"; - int rc = 0; - ENTRY; - - cfs_daemonize(name); - - 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; + int rc = 0; + bool first = true; + ENTRY; - /* Any new or requeued lostlocks will change the state */ - rq_state &= ~(RQ_NOW | RQ_LATER); - cfs_spin_unlock(&config_list_lock); + CDEBUG(D_MGC, "Starting requeue thread\n"); + + /* Keep trying failed locks periodically */ + 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); + + if (first) { + first = false; + complete(&rq_start); + } - /* Always wait a few seconds to allow the server who - caused the lock revocation to finish its setup, plus some - random so everyone doesn't try to reconnect at once. */ - to = MGC_TIMEOUT_MIN_SECONDS * CFS_HZ; - to += rand * CFS_HZ / 100; /* rand is centi-seconds */ - lwi = LWI_TIMEOUT(to, NULL, NULL); - l_wait_event(rq_waitq, rq_state & RQ_STOP, &lwi); + /* Always wait a few seconds to allow the server who + caused the lock revocation to finish its setup, plus some + random so everyone doesn't try to reconnect at once. */ + to = msecs_to_jiffies(MGC_TIMEOUT_MIN_SECONDS * MSEC_PER_SEC); + /* rand is centi-seconds */ + to += msecs_to_jiffies(rand * MSEC_PER_SEC / 100); + lwi = LWI_TIMEOUT(to, NULL, NULL); + l_wait_event(rq_waitq, rq_state & (RQ_STOP | RQ_PRECLEANUP), + &lwi); /* * iterate & processing through the list. for each cld, process @@ -514,226 +625,291 @@ 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); + rq_state &= ~RQ_PRECLEANUP; + 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); + LASSERT(atomic_read(&cld->cld_refcount) > 0); - /* Whether we enqueued again or not in mgc_process_log, - * we're done with the ref from the old enqueue */ - if (cld_prev) - config_log_put(cld_prev); - cld_prev = cld; + /* Whether we enqueued again or not in mgc_process_log, + * we're done with the ref from the old enqueue */ + if (cld_prev) + config_log_put(cld_prev); + cld_prev = cld; cld->cld_lostlock = 0; - if (likely(!stopped)) - do_requeue(cld); + 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; - } + 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); - 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); + /* 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); - cfs_complete(&rq_exit); + complete(&rq_exit); - CDEBUG(D_MGC, "Ending requeue thread\n"); - RETURN(rc); + CDEBUG(D_MGC, "Ending requeue thread\n"); + RETURN(rc); } /* Add a cld to the list to requeue. Start the requeue thread if needed. We are responsible for dropping the config log reference from here on out. */ static void mgc_requeue_add(struct config_llog_data *cld) { - ENTRY; + ENTRY; - CDEBUG(D_INFO, "log %s: requeue (r=%d sp=%d st=%x)\n", - cld->cld_logname, cfs_atomic_read(&cld->cld_refcount), - cld->cld_stopping, rq_state); - LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0); + CDEBUG(D_INFO, "log %s: requeue (r=%d sp=%d st=%x)\n", + cld->cld_logname, atomic_read(&cld->cld_refcount), + cld->cld_stopping, rq_state); + LASSERT(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); + wake_up(&rq_waitq); + } + EXIT; } /********************** class fns **********************/ +static int mgc_local_llog_init(const struct lu_env *env, + struct obd_device *obd, + struct obd_device *disk) +{ + struct llog_ctxt *ctxt; + int rc; + + ENTRY; + + rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, disk, + &llog_osd_ops); + if (rc) + RETURN(rc); -static int mgc_fs_setup(struct obd_device *obd, struct super_block *sb, - struct vfsmount *mnt) + ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); + LASSERT(ctxt); + ctxt->loc_dir = obd->u.cli.cl_mgc_configs_dir; + llog_ctxt_put(ctxt); + + RETURN(0); +} + +static int mgc_local_llog_fini(const struct lu_env *env, + struct obd_device *obd) { - struct lvfs_run_ctxt saved; - struct lustre_sb_info *lsi = s2lsi(sb); - struct client_obd *cli = &obd->u.cli; - struct dentry *dentry; - char *label; - int err = 0; - ENTRY; + struct llog_ctxt *ctxt; - LASSERT(lsi); - LASSERT(lsi->lsi_srv_mnt == mnt); + ENTRY; - /* The mgc fs exclusion sem. Only one fs can be setup at a time. */ - cfs_down(&cli->cl_mgc_sem); + ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); + llog_cleanup(env, ctxt); - cfs_cleanup_group_info(); + RETURN(0); +} - 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", lsi->lsi_fstype, - PTR_ERR(obd->obd_fsops)); - RETURN(PTR_ERR(obd->obd_fsops)); - } +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; + int rc = 0; - cli->cl_mgc_vfsmnt = mnt; - err = fsfilt_setup(obd, mnt->mnt_sb); - if (err) - GOTO(err_ops, err); - - OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt); - obd->obd_lvfs_ctxt.pwdmnt = mnt; - obd->obd_lvfs_ctxt.pwd = mnt->mnt_root; - obd->obd_lvfs_ctxt.fs = get_ds(); - - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - dentry = ll_lookup_one_len(MOUNT_CONFIGS_DIR, cfs_fs_pwd(current->fs), - strlen(MOUNT_CONFIGS_DIR)); - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - if (IS_ERR(dentry)) { - err = PTR_ERR(dentry); - CERROR("cannot lookup %s directory: rc = %d\n", - MOUNT_CONFIGS_DIR, err); - GOTO(err_ops, err); - } - cli->cl_mgc_configs_dir = dentry; + ENTRY; + + LASSERT(lsi); + LASSERT(lsi->lsi_dt_dev); + + /* The mgc fs exclusion mutex. Only one fs can be setup at a time. */ + mutex_lock(&cli->cl_mgc_mutex); + + /* Setup the configs dir */ + 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) + RETURN(rc); + + rc = dt_root_get(env, lsi->lsi_dt_dev, &rfid); + if (rc) + GOTO(out_los, rc); - /* We take an obd ref to insure that we can't get to mgc_cleanup - without calling mgc_fs_cleanup first. */ - class_incref(obd, "mgc_fs", obd); + root = dt_locate_at(env, lsi->lsi_dt_dev, &rfid, + &cli->cl_mgc_los->los_dev->dd_lu_dev, NULL); + if (unlikely(IS_ERR(root))) + GOTO(out_los, rc = PTR_ERR(root)); - label = fsfilt_get_label(obd, mnt->mnt_sb); - if (label) - CDEBUG(D_MGC, "MGC using disk labelled=%s\n", label); + dto = local_file_find_or_create(env, cli->cl_mgc_los, root, + MOUNT_CONFIGS_DIR, + S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO); + lu_object_put_nocache(env, &root->do_lu); + if (IS_ERR(dto)) + GOTO(out_los, rc = PTR_ERR(dto)); - /* We keep the cl_mgc_sem until mgc_fs_cleanup */ - RETURN(0); + cli->cl_mgc_configs_dir = dto; -err_ops: - fsfilt_put_ops(obd->obd_fsops); - obd->obd_fsops = NULL; - cli->cl_mgc_vfsmnt = NULL; - cfs_up(&cli->cl_mgc_sem); - RETURN(err); + LASSERT(lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt); + rc = mgc_local_llog_init(env, obd, lsi->lsi_osd_exp->exp_obd); + if (rc) + GOTO(out_llog, rc); + + /* We take an obd ref to insure that we can't get to mgc_cleanup + * without calling mgc_fs_cleanup first. */ + class_incref(obd, "mgc_fs", obd); + + /* We keep the cl_mgc_sem until mgc_fs_cleanup */ + EXIT; +out_llog: + if (rc) { + lu_object_put(env, &cli->cl_mgc_configs_dir->do_lu); + cli->cl_mgc_configs_dir = NULL; + } +out_los: + if (rc < 0) { + local_oid_storage_fini(env, cli->cl_mgc_los); + cli->cl_mgc_los = NULL; + mutex_unlock(&cli->cl_mgc_mutex); + } + 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 client_obd *cli = &obd->u.cli; - int rc = 0; - ENTRY; + struct client_obd *cli = &obd->u.cli; + ENTRY; - LASSERT(cli->cl_mgc_vfsmnt != NULL); + LASSERT(cli->cl_mgc_los != NULL); - if (cli->cl_mgc_configs_dir != NULL) { - struct lvfs_run_ctxt saved; - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - l_dput(cli->cl_mgc_configs_dir); - cli->cl_mgc_configs_dir = NULL; - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - class_decref(obd, "mgc_fs", obd); - } + mgc_local_llog_fini(env, obd); - cli->cl_mgc_vfsmnt = NULL; - if (obd->obd_fsops) - fsfilt_put_ops(obd->obd_fsops); + lu_object_put_nocache(env, &cli->cl_mgc_configs_dir->do_lu); + cli->cl_mgc_configs_dir = NULL; - cfs_up(&cli->cl_mgc_sem); + local_oid_storage_fini(env, cli->cl_mgc_los); + cli->cl_mgc_los = NULL; - RETURN(rc); + class_decref(obd, "mgc_fs", obd); + mutex_unlock(&cli->cl_mgc_mutex); + + RETURN(0); } -static cfs_atomic_t mgc_count = CFS_ATOMIC_INIT(0); -static int mgc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) +static int mgc_llog_init(const struct lu_env *env, struct obd_device *obd) { - int rc = 0; - ENTRY; + struct llog_ctxt *ctxt; + int rc; - switch (stage) { - case OBD_CLEANUP_EARLY: - break; - case OBD_CLEANUP_EXPORTS: - 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); - } - } - obd_cleanup_client_import(obd); - rc = obd_llog_finish(obd, 0); - if (rc != 0) - CERROR("failed to cleanup llogging subsystems\n"); - break; - } - RETURN(rc); + ENTRY; + + /* setup only remote ctxt, the local disk context is switched per each + * filesystem during mgc_fs_setup() */ + rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CONFIG_REPL_CTXT, obd, + &llog_client_ops); + if (rc) + RETURN(rc); + + ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT); + LASSERT(ctxt); + + llog_initiator_connect(ctxt); + llog_ctxt_put(ctxt); + + RETURN(0); +} + +static int mgc_llog_fini(const struct lu_env *env, struct obd_device *obd) +{ + struct llog_ctxt *ctxt; + + ENTRY; + + ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT); + if (ctxt) + llog_cleanup(env, ctxt); + + RETURN(0); +} + + +static atomic_t mgc_count = ATOMIC_INIT(0); +static int mgc_precleanup(struct obd_device *obd) +{ + int rc = 0; + int temp; + ENTRY; + + 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); } static int mgc_cleanup(struct obd_device *obd) { - struct client_obd *cli = &obd->u.cli; int rc; ENTRY; - LASSERT(cli->cl_mgc_vfsmnt == NULL); - /* COMPAT_146 - old config logs may have added profiles we don't know about */ if (obd->obd_type->typ_refcnt <= 1) @@ -749,50 +925,55 @@ static int mgc_cleanup(struct obd_device *obd) static int mgc_setup(struct obd_device *obd, struct lustre_cfg *lcfg) { - struct lprocfs_static_vars lvars; - int rc; - ENTRY; - - ptlrpcd_addref(); - - rc = client_obd_setup(obd, lcfg); - if (rc) - GOTO(err_decref, rc); + struct task_struct *task; + int rc; + ENTRY; - rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL); - if (rc) { - CERROR("failed to setup llogging subsystems\n"); - GOTO(err_cleanup, rc); - } + rc = ptlrpcd_addref(); + if (rc < 0) + RETURN(rc); - lprocfs_mgc_init_vars(&lvars); - lprocfs_obd_setup(obd, lvars.obd_vars); - sptlrpc_lprocfs_cliobd_attach(obd); + rc = client_obd_setup(obd, lcfg); + if (rc) + GOTO(err_decref, rc); - if (cfs_atomic_inc_return(&mgc_count) == 1) { - rq_state = 0; - cfs_waitq_init(&rq_waitq); + rc = mgc_llog_init(NULL, obd); + if (rc) { + CERROR("failed to setup llogging subsystems\n"); + GOTO(err_cleanup, rc); + } - /* start requeue thread */ - rc = cfs_create_thread(mgc_requeue_thread, NULL, - CFS_DAEMON_FLAGS); - if (rc < 0) { - CERROR("%s: Cannot start requeue thread (%d)," - "no more log updates!\n", - obd->obd_name, rc); - GOTO(err_cleanup, rc); - } - /* rc is the pid of mgc_requeue_thread. */ - rc = 0; - } +#ifdef CONFIG_PROC_FS + obd->obd_vars = lprocfs_mgc_obd_vars; + lprocfs_obd_setup(obd); +#endif + sptlrpc_lprocfs_cliobd_attach(obd); + + if (atomic_inc_return(&mgc_count) == 1) { + rq_state = 0; + init_waitqueue_head(&rq_waitq); + + /* start requeue thread */ + task = kthread_run(mgc_requeue_thread, NULL, "ll_cfg_requeue"); + if (IS_ERR(task)) { + rc = PTR_ERR(task); + CERROR("%s: cannot start requeue thread: rc = %d; " + "no more log updates\n", + obd->obd_name, rc); + GOTO(err_cleanup, rc); + } + /* rc is the task_struct pointer of mgc_requeue_thread. */ + rc = 0; + wait_for_completion(&rq_start); + } - RETURN(rc); + RETURN(rc); err_cleanup: - client_obd_cleanup(obd); + client_obd_cleanup(obd); err_decref: - ptlrpcd_decref(); - RETURN(rc); + ptlrpcd_decref(); + RETURN(rc); } /* based on ll_mdc_blocking_ast */ @@ -809,39 +990,39 @@ static int mgc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, /* mgs wants the lock, give it up... */ LDLM_DEBUG(lock, "MGC blocking CB"); ldlm_lock2handle(lock, &lockh); - rc = ldlm_cli_cancel(&lockh); - break; - case LDLM_CB_CANCELING: - /* We've given up the lock, prepare ourselves to update. */ - LDLM_DEBUG(lock, "MGC cancel CB"); - - CDEBUG(D_MGC, "Lock res "LPX64" (%.8s)\n", - lock->l_resource->lr_name.name[0], - (char *)&lock->l_resource->lr_name.name[0]); - - if (!cld) { - CDEBUG(D_INFO, "missing data, won't requeue\n"); - break; - } + rc = ldlm_cli_cancel(&lockh, LCF_ASYNC); + break; + case LDLM_CB_CANCELING: + /* We've given up the lock, prepare ourselves to update. */ + LDLM_DEBUG(lock, "MGC cancel CB"); + + CDEBUG(D_MGC, "Lock res "DLDLMRES" (%.8s)\n", + PLDLMRES(lock->l_resource), + (char *)&lock->l_resource->lr_name.name[0]); + + if (!cld) { + CDEBUG(D_INFO, "missing data, won't requeue\n"); + break; + } - /* held at mgc_process_log(). */ - LASSERT(cfs_atomic_read(&cld->cld_refcount) > 0); - /* Are we done with this log? */ - if (cld->cld_stopping) { - CDEBUG(D_MGC, "log %s: stopping, won't requeue\n", - cld->cld_logname); - config_log_put(cld); - break; - } - /* 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) { - CDEBUG(D_MGC, "log %.8s: disconnecting, won't requeue\n", - cld->cld_logname); - config_log_put(cld); - break; - } + /* held at mgc_process_log(). */ + LASSERT(atomic_read(&cld->cld_refcount) > 0); + /* Are we done with this log? */ + if (cld->cld_stopping) { + CDEBUG(D_MGC, "log %s: stopping, won't requeue\n", + cld->cld_logname); + config_log_put(cld); + break; + } + /* Make sure not to re-enqueue when the mgc is stopping + (we get called from client_disconnect_export) */ + 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); + break; + } /* Re-enqueue now */ mgc_requeue_add(cld); @@ -855,7 +1036,10 @@ static int mgc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, } /* Not sure where this should go... */ -#define MGC_ENQUEUE_LIMIT 50 +/* This is the timeout value for MGS_CONNECT request plus a ping interval, such + * that we can have a chance to try the secondary MGS if any. */ +#define MGC_ENQUEUE_LIMIT (INITIAL_CONNECT_TIMEOUT + (AT_OFF ? 0 : at_min) \ + + PING_INTERVAL) #define MGC_TARGET_REG_LIMIT 10 #define MGC_SEND_PARAM_LIMIT 10 @@ -897,19 +1081,23 @@ 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) +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) { - struct config_llog_data *cld = (struct config_llog_data *)data; - struct ldlm_enqueue_info einfo = { type, mode, mgc_blocking_ast, - ldlm_completion_ast, NULL, NULL, NULL }; - struct ptlrpc_request *req; - int short_limit = cld_is_sptlrpc(cld); - int rc; - ENTRY; + struct config_llog_data *cld = (struct config_llog_data *)data; + struct ldlm_enqueue_info einfo = { + .ei_type = type, + .ei_mode = mode, + .ei_cb_bl = mgc_blocking_ast, + .ei_cb_cp = ldlm_completion_ast, + }; + struct ptlrpc_request *req; + int short_limit = cld_is_sptlrpc(cld); + int rc; + ENTRY; CDEBUG(D_MGC, "Enqueue for %s (res "LPX64")\n", cld->cld_logname, cld->cld_resid.name[0]); @@ -921,6 +1109,8 @@ 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 */ @@ -932,32 +1122,32 @@ static int mgc_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm, /* 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); RETURN(rc); } -static int mgc_cancel(struct obd_export *exp, struct lov_stripe_md *md, - __u32 mode, struct lustre_handle *lockh) +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) { - /* 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); + wake_up(&rq_waitq); - /* TODO: Help the MGS rebuild nidtbl. -jay */ + /* TODO: Help the MGS rebuild nidtbl. -jay */ } /* Send target_reg message to MGS */ @@ -1000,9 +1190,10 @@ static int mgc_target_register(struct obd_export *exp, RETURN(rc); } -int mgc_set_info_async(const struct lu_env *env, struct obd_export *exp, - obd_count keylen, void *key, obd_count vallen, - void *val, struct ptlrpc_request_set *set) +static int mgc_set_info_async(const struct lu_env *env, struct obd_export *exp, + u32 keylen, void *key, + u32 vallen, void *val, + struct ptlrpc_request_set *set) { int rc = -EINVAL; ENTRY; @@ -1036,27 +1227,21 @@ int mgc_set_info_async(const struct lu_env *env, struct obd_export *exp, rc = mgc_target_register(exp, mti); RETURN(rc); } - if (KEY_IS(KEY_SET_FS)) { - struct super_block *sb = (struct super_block *)val; - struct lustre_sb_info *lsi; - if (vallen != sizeof(struct super_block)) - RETURN(-EINVAL); - lsi = s2lsi(sb); - rc = mgc_fs_setup(exp->exp_obd, sb, lsi->lsi_srv_mnt); - if (rc) { - CERROR("set_fs got %d\n", rc); - } - RETURN(rc); - } - if (KEY_IS(KEY_CLEAR_FS)) { - if (vallen != 0) - RETURN(-EINVAL); - rc = mgc_fs_cleanup(exp->exp_obd); - if (rc) { - CERROR("clear_fs got %d\n", rc); - } - RETURN(rc); - } + if (KEY_IS(KEY_SET_FS)) { + struct super_block *sb = (struct super_block *)val; + + if (vallen != sizeof(struct super_block)) + RETURN(-EINVAL); + + 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(env, exp->exp_obd); + RETURN(rc); + } if (KEY_IS(KEY_SET_INFO)) { struct mgs_send_param *msp; @@ -1112,8 +1297,7 @@ 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; @@ -1142,6 +1326,8 @@ static int mgc_import_event(struct obd_device *obd, switch (event) { case IMP_EVENT_DISCON: /* MGC imports should not wait for recovery */ + if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV)) + ptlrpc_pinger_ir_down(); break; case IMP_EVENT_INACTIVE: break; @@ -1150,12 +1336,14 @@ 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); + if (OCD_HAS_FLAG(&imp->imp_connect_data, IMP_RECOV)) + ptlrpc_pinger_ir_up(); + break; case IMP_EVENT_OCD: break; case IMP_EVENT_DEACTIVATE: @@ -1168,58 +1356,8 @@ static int mgc_import_event(struct obd_device *obd, RETURN(rc); } -static int mgc_llog_init(struct obd_device *obd, struct obd_llog_group *olg, - struct obd_device *tgt, int *index) -{ - struct llog_ctxt *ctxt; - int rc; - ENTRY; - - LASSERT(olg == &obd->obd_olg); - - rc = llog_setup(NULL, obd, olg, LLOG_CONFIG_ORIG_CTXT, tgt, - &llog_lvfs_ops); - if (rc) - RETURN(rc); - - 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; - - ENTRY; - - ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT); - if (ctxt) - llog_cleanup(NULL, ctxt); - - ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); - if (ctxt) - llog_cleanup(NULL, ctxt); - RETURN(0); -} - enum { - CONFIG_READ_NRPAGES_INIT = 1 << (20 - CFS_PAGE_SHIFT), + CONFIG_READ_NRPAGES_INIT = 1 << (20 - PAGE_CACHE_SHIFT), CONFIG_READ_NRPAGES = 4 }; @@ -1245,27 +1383,34 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, LASSERT(cfg->cfg_instance != NULL); LASSERT(cfg->cfg_sb == cfg->cfg_instance); - OBD_ALLOC(inst, CFS_PAGE_SIZE); - if (inst == NULL) - RETURN(-ENOMEM); + OBD_ALLOC(inst, PAGE_CACHE_SIZE); + if (inst == NULL) + RETURN(-ENOMEM); if (!IS_SERVER(lsi)) { - pos = sprintf(inst, "%p", cfg->cfg_instance); + pos = snprintf(inst, PAGE_CACHE_SIZE, "%p", cfg->cfg_instance); + if (pos >= PAGE_CACHE_SIZE) { + OBD_FREE(inst, PAGE_CACHE_SIZE); + return -E2BIG; + } } else { LASSERT(IS_MDT(lsi)); - rc = server_name2svname(lsi->lsi_svname, inst, NULL); - if (rc) + rc = server_name2svname(lsi->lsi_svname, inst, NULL, + PAGE_CACHE_SIZE); + if (rc) { + OBD_FREE(inst, PAGE_CACHE_SIZE); RETURN(-EINVAL); + } pos = strlen(inst); } ++pos; buf = inst + pos; - bufsz = CFS_PAGE_SIZE - pos; + bufsz = PAGE_CACHE_SIZE - pos; while (datalen > 0) { int entry_len = sizeof(*entry); - int is_ost; + int is_ost, i; struct obd_device *obd; char *obdname; char *cname; @@ -1293,7 +1438,7 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, /* 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) { + if (entry->mne_length > PAGE_CACHE_SIZE) { CERROR("MNE too large (%u)\n", entry->mne_length); break; } @@ -1363,11 +1508,26 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, pos += sprintf(params, "%s.import=%s", cname, "connection="); uuid = buf + pos; - /* 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); + 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; + } + + /* iterate all nids to find one */ + /* find uuid by nid */ + rc = -ENOENT; + for (i = 0; i < entry->mne_nid_count; i++) { + rc = client_import_find_conn(obd->u.cli.cl_import, + entry->u.nids[i], + (struct obd_uuid *)uuid); + if (rc == 0) + break; + } + + 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])); @@ -1383,12 +1543,11 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, lustre_cfg_bufs_set_string(&bufs, 1, params); - rc = -ENOMEM; - lcfg = lustre_cfg_new(LCFG_PARAM, &bufs); - if (lcfg == NULL) { - CERROR("mgc: cannot allocate memory\n"); - break; - } + lcfg = lustre_cfg_new(LCFG_PARAM, &bufs); + if (lcfg == NULL) { + rc = -ENOMEM; + break; + } CDEBUG(D_INFO, "ir apply logs "LPD64"/"LPD64" for %s -> %s\n", prev_version, max_version, obdname, params); @@ -1402,30 +1561,35 @@ static int mgc_apply_recover_logs(struct obd_device *mgc, /* continue, even one with error */ } - OBD_FREE(inst, CFS_PAGE_SIZE); + OBD_FREE(inst, PAGE_CACHE_SIZE); RETURN(rc); } /** * 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; - cfs_page_t **pages; - int nrpages; - bool eof = true; + 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; + __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, @@ -1434,7 +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); @@ -1442,418 +1606,488 @@ static int mgc_process_recover_log(struct obd_device *obd, GOTO(out, rc = -ENOMEM); for (i = 0; i < nrpages; i++) { - pages[i] = cfs_alloc_page(CFS_ALLOC_STD); + pages[i] = alloc_page(GFP_KERNEL); if (pages[i] == NULL) GOTO(out, rc = -ENOMEM); } again: - LASSERT(cld_is_recover(cld)); - LASSERT(cfs_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); - - 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)); - strncpy(body->mcb_name, cld->cld_logname, sizeof(body->mcb_name)); - body->mcb_offset = cfg->cfg_last_idx + 1; - body->mcb_type = cld->cld_type; - body->mcb_bits = CFS_PAGE_SHIFT; - body->mcb_units = nrpages; - - /* allocate bulk transfer descriptor */ - desc = ptlrpc_prep_bulk_imp(req, nrpages, BULK_PUT_SINK, - MGS_BULK_PORTAL); - if (desc == NULL) - GOTO(out, rc = -ENOMEM); - - for (i = 0; i < nrpages; i++) - ptlrpc_prep_bulk_page(desc, pages[i], 0, CFS_PAGE_SIZE); +#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); + + rc = ptlrpc_request_pack(req, LUSTRE_MGS_VERSION, MGS_CONFIG_READ); + if (rc) + GOTO(out, rc); - ptlrpc_request_set_replen(req); - rc = ptlrpc_queue_wait(req); - 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)); + if (strlcpy(body->mcb_name, cld->cld_logname, sizeof(body->mcb_name)) + >= sizeof(body->mcb_name)) + GOTO(out, rc = -E2BIG); + 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; + + /* allocate bulk transfer descriptor */ + 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++) + 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); - 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); + + 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); + } - /* 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; + ealen = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, 0); + if (ealen < 0) + GOTO(out, rc = ealen); - CDEBUG(D_INFO, "Latest version "LPD64", more %d.\n", - res->mcr_offset, eof == false); + if (ealen > nrpages << PAGE_CACHE_SHIFT) + GOTO(out, rc = -EINVAL); - ealen = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, 0); - if (ealen < 0) - GOTO(out, rc = ealen); + 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; - if (ealen > nrpages << CFS_PAGE_SHIFT) - GOTO(out, rc = -EINVAL); + CDEBUG(D_INFO, "nodemap config changed in transit, retrying\n"); - if (ealen == 0) { /* no logs transferred */ - if (!eof) - rc = -EINVAL; - GOTO(out, rc); - } + /* setting eof to false, we request config again */ + eof = false; + GOTO(out, rc = 0); + } +#endif + if (!eof) + rc = -EINVAL; + GOTO(out, rc); + } mne_swab = !!ptlrpc_rep_need_swab(req); -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0) +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 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), - mne_swab); - cfs_kunmap(pages[i]); - if (rc2 < 0) { - CWARN("Process recover log %s error %d\n", - cld->cld_logname, rc2); - break; - } + /* 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; + union lu_page *ptr; + + ptr = kmap(pages[i]); + 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("%s: error processing %s log %s: rc = %d\n", + obd->obd_name, + cld_is_nodemap(cld) ? "nodemap" : "recovery", + cld->cld_logname, + rc2); + break; + } - ealen -= CFS_PAGE_SIZE; - } + ealen -= PAGE_CACHE_SIZE; + } out: - if (req) - ptlrpc_req_finished(req); - - if (rc == 0 && !eof) - goto again; - - if (pages) { - for (i = 0; i < nrpages; i++) { - if (pages[i] == NULL) - break; - cfs_free_page(pages[i]); - } - OBD_FREE(pages, sizeof(*pages) * nrpages); - } - 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; - - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - rc = llog_open(NULL, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS); - if (rc == 0) { - llog_init_handle(NULL, llh, LLOG_F_IS_PLAIN, NULL); - rc = llog_get_size(llh); - llog_close(NULL, llh); - } else if (rc == -ENOENT) { - rc = 0; + if (req) { + ptlrpc_req_finished(req); + req = NULL; } - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - /* header is record 1 */ - return (rc <= 1); -} -static int mgc_copy_handler(const struct lu_env *env, struct llog_handle *llh, - struct llog_rec_hdr *rec, void *data) -{ - 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(env, local_llh, &local_rec, NULL, 0, - (void *)cfg_buf, -1); + if (rc == 0 && !eof) + goto again; - 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)); +#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 - RETURN(rc); + if (pages) { + for (i = 0; i < nrpages; i++) { + if (pages[i] == NULL) + break; + __free_page(pages[i]); + } + OBD_FREE(pages, sizeof(*pages) * nrpages); + } + return rc; } /* Copy a remote log locally */ -static int mgc_copy_llog(struct obd_device *obd, struct llog_ctxt *rctxt, - struct llog_ctxt *lctxt, char *logname) +static int mgc_llog_local_copy(const struct lu_env *env, + struct obd_device *obd, + struct llog_ctxt *rctxt, + struct llog_ctxt *lctxt, char *logname) { - struct llog_handle *local_llh, *remote_llh; - struct obd_uuid *uuid; - char *temp_log; - int rc, rc2; - ENTRY; + char *temp_log; + int rc; - /* Write new log to a temp name, then vfs_rename over logname - upon successful completion. */ + ENTRY; - OBD_ALLOC(temp_log, strlen(logname) + 1); - if (!temp_log) - RETURN(-ENOMEM); - sprintf(temp_log, "%sT", logname); + /* + * - copy it to backup using llog_backup() + * - copy remote llog to logname using llog_backup() + * - if failed then move bakup to logname again + */ - /* Make sure there's no old temp log */ - rc = llog_erase(NULL, lctxt, NULL, temp_log); - if (rc < 0 && rc != -ENOENT) - GOTO(out, rc); + OBD_ALLOC(temp_log, strlen(logname) + 2); + if (!temp_log) + RETURN(-ENOMEM); + sprintf(temp_log, "%sT", logname); - /* open local log */ - rc = llog_open_create(NULL, lctxt, &local_llh, NULL, temp_log); - if (rc) + /* make a copy of local llog at first */ + rc = llog_backup(env, obd, lctxt, lctxt, logname, temp_log); + if (rc < 0 && rc != -ENOENT) 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, mgc_copy_handler, - (void *)local_llh, NULL); - -out_closer: - rc2 = llog_close(NULL, remote_llh); - if (!rc) - rc = rc2; -out_closel: - 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 == 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, - temp_log, logname); - } - CDEBUG(D_MGC, "Copied remote log %s (%d)\n", logname, rc); + /* copy remote llog to the local copy */ + rc = llog_backup(env, obd, rctxt, lctxt, logname, logname); + if (rc == -ENOENT) { + /* no remote llog, delete local one too */ + llog_erase(env, lctxt, NULL, logname); + } else if (rc < 0) { + /* error during backup, get local one back from the copy */ + llog_backup(env, obd, lctxt, lctxt, temp_log, logname); out: - if (rc) - CERROR("Failed to copy remote log %s (%d)\n", logname, rc); - OBD_FREE(temp_log, strlen(logname) + 1); - RETURN(rc); + CERROR("%s: failed to copy remote log %s: rc = %d\n", + obd->obd_name, logname, rc); + } + llog_erase(env, lctxt, NULL, temp_log); + OBD_FREE(temp_log, strlen(logname) + 2); + return rc; } /* local_only means it cannot get remote llogs */ static int mgc_process_cfg_log(struct obd_device *mgc, - struct config_llog_data *cld, - int local_only) + struct config_llog_data *cld, int local_only) { - struct llog_ctxt *ctxt, *lctxt = NULL; - struct client_obd *cli = &mgc->u.cli; - struct lvfs_run_ctxt *saved_ctxt; - struct lustre_sb_info *lsi = NULL; - int rc = 0, must_pop = 0; - bool sptlrpc_started = false; - - ENTRY; + struct llog_ctxt *ctxt, *lctxt = NULL; + struct client_obd *cli = &mgc->u.cli; + struct lustre_sb_info *lsi = NULL; + int rc = 0; + bool sptlrpc_started = false; + struct lu_env *env; - LASSERT(cld); - LASSERT(cfs_mutex_is_locked(&cld->cld_lock)); + ENTRY; - /* - * local copy of sptlrpc log is controlled elsewhere, don't try to - * read it up here. - */ - if (cld_is_sptlrpc(cld) && local_only) - RETURN(0); + LASSERT(cld); + LASSERT(mutex_is_locked(&cld->cld_lock)); - if (cld->cld_cfg.cfg_sb) - lsi = s2lsi(cld->cld_cfg.cfg_sb); + /* + * local copy of sptlrpc log is controlled elsewhere, don't try to + * read it up here. + */ + if (cld_is_sptlrpc(cld) && local_only) + RETURN(0); - ctxt = llog_get_context(mgc, LLOG_CONFIG_REPL_CTXT); - if (!ctxt) { - CERROR("missing llog context\n"); - RETURN(-EINVAL); - } + if (cld->cld_cfg.cfg_sb) + lsi = s2lsi(cld->cld_cfg.cfg_sb); - OBD_ALLOC_PTR(saved_ctxt); - if (saved_ctxt == NULL) - RETURN(-ENOMEM); + OBD_ALLOC_PTR(env); + if (env == NULL) + RETURN(-ENOMEM); - lctxt = llog_get_context(mgc, LLOG_CONFIG_ORIG_CTXT); - - /* 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 && IS_SERVER(lsi) && - (lsi->lsi_srv_mnt == cli->cl_mgc_vfsmnt) && - !IS_MGS(lsi) && lsi->lsi_srv_mnt) { - push_ctxt(saved_ctxt, &mgc->obd_lvfs_ctxt, NULL); - must_pop++; - if (!local_only) - /* Only try to copy log if we have the lock. */ - rc = mgc_copy_llog(mgc, ctxt, lctxt, cld->cld_logname); - if (local_only || rc) { - if (mgc_llog_is_empty(mgc, lctxt, cld->cld_logname)) { - LCONSOLE_ERROR_MSG(0x13a, "Failed to get MGS " - "log %s and no local copy." - "\n", cld->cld_logname); - GOTO(out_pop, rc = -ENOTCONN); - } - CDEBUG(D_MGC, "Failed to get MGS log %s, using local " - "copy for now, will try to update later.\n", - cld->cld_logname); - } - /* Now, whether we copied or not, start using the local llog. - If we failed to copy, we'll start using whatever the old - log has. */ - llog_ctxt_put(ctxt); - ctxt = lctxt; - lctxt = NULL; - } else if (local_only) { /* no local log at client side */ - GOTO(out_pop, rc = -EIO); - } + rc = lu_env_init(env, LCT_MG_THREAD); + if (rc) + GOTO(out_free, rc); + + ctxt = llog_get_context(mgc, LLOG_CONFIG_REPL_CTXT); + LASSERT(ctxt); + + lctxt = llog_get_context(mgc, LLOG_CONFIG_ORIG_CTXT); + + /* 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 && IS_SERVER(lsi) && !IS_MGS(lsi) && + cli->cl_mgc_configs_dir != NULL && + lu2dt_dev(cli->cl_mgc_configs_dir->do_lu.lo_dev) == + lsi->lsi_dt_dev) { + if (!local_only) + /* Only try to copy log if we have the lock. */ + rc = mgc_llog_local_copy(env, mgc, ctxt, lctxt, + cld->cld_logname); + if (local_only || rc) { + if (strcmp(cld->cld_logname, PARAMS_FILENAME) != 0 && + llog_is_empty(env, lctxt, cld->cld_logname)) { + LCONSOLE_ERROR_MSG(0x13a, "Failed to get MGS " + "log %s and no local copy." + "\n", cld->cld_logname); + GOTO(out_pop, rc = -ENOENT); + } + CDEBUG(D_MGC, "Failed to get MGS log %s, using local " + "copy for now, will try to update later.\n", + cld->cld_logname); + rc = 0; + } + /* Now, whether we copied or not, start using the local llog. + * If we failed to copy, we'll start using whatever the old + * log has. */ + llog_ctxt_put(ctxt); + ctxt = lctxt; + lctxt = NULL; + } else { + if (local_only) /* no local log at client side */ + GOTO(out_pop, rc = -EIO); + } - if (cld_is_sptlrpc(cld)) { - sptlrpc_conf_log_update_begin(cld->cld_logname); - sptlrpc_started = true; - } + if (cld_is_sptlrpc(cld)) { + sptlrpc_conf_log_update_begin(cld->cld_logname); + sptlrpc_started = true; + } - /* 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); - EXIT; + /* 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(env, ctxt, cld->cld_logname, + &cld->cld_cfg); + EXIT; out_pop: - llog_ctxt_put(ctxt); - if (lctxt) - llog_ctxt_put(lctxt); - if (must_pop) - pop_ctxt(saved_ctxt, &mgc->obd_lvfs_ctxt, NULL); - - OBD_FREE_PTR(saved_ctxt); - /* - * update settings on existing OBDs. doing it inside - * of llog_process_lock so no device is attaching/detaching - * in parallel. - * the logname must be -sptlrpc - */ - if (sptlrpc_started) { - LASSERT(cld_is_sptlrpc(cld)); - sptlrpc_conf_log_update_end(cld->cld_logname); - class_notify_sptlrpc_conf(cld->cld_logname, - strlen(cld->cld_logname) - - strlen("-sptlrpc")); - } + __llog_ctxt_put(env, ctxt); + if (lctxt) + __llog_ctxt_put(env, lctxt); + + /* + * update settings on existing OBDs. doing it inside + * of llog_process_lock so no device is attaching/detaching + * in parallel. + * the logname must be -sptlrpc + */ + if (sptlrpc_started) { + LASSERT(cld_is_sptlrpc(cld)); + sptlrpc_conf_log_update_end(cld->cld_logname); + class_notify_sptlrpc_conf(cld->cld_logname, + strlen(cld->cld_logname) - + strlen("-sptlrpc")); + } - RETURN(rc); + lu_env_fini(env); +out_free: + OBD_FREE_PTR(env); + return rc; +} + +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 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) +/** + * 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 }; - int rc = 0, rcl, flags = LDLM_FL_NO_LRU; + __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.) */ - cfs_mutex_lock(&cld->cld_lock); - if (cld->cld_stopping) { - cfs_mutex_unlock(&cld->cld_lock); +restart: + mutex_lock(&cld->cld_lock); + if (cld->cld_stopping) { + mutex_unlock(&cld->cld_lock); RETURN(0); } 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); - cfs_mutex_unlock(&cld->cld_lock); + mutex_unlock(&cld->cld_lock); - /* Now drop the lock so MGS can revoke it */ - if (!rcl) { - rcl = mgc_cancel(mgc->u.cli.cl_mgc_mgsexp, NULL, - LCK_CR, &lockh); - if (rcl) - CERROR("Can't drop cfg lock: %d\n", rcl); - } + /* Now drop the lock so MGS can revoke it */ + if (!rcl) { + rcl = mgc_cancel(mgc->u.cli.cl_mgc_mgsexp, LCK_CR, &lockh); + if (rcl) + CERROR("Can't drop cfg lock: %d\n", rcl); + } - RETURN(rc); + RETURN(rc); } @@ -1861,7 +2095,7 @@ int mgc_process_log(struct obd_device *mgc, struct config_llog_data *cld) * LCFG_LOG_START gets the config log from the MGS, processes it to start * any services, and adds it to the list logs to watch (follow). */ -static int mgc_process_config(struct obd_device *obd, obd_count len, void *buf) +static int mgc_process_config(struct obd_device *obd, size_t len, void *buf) { struct lustre_cfg *lcfg = buf; struct config_llog_instance *cfg = NULL; @@ -1932,6 +2166,19 @@ static int mgc_process_config(struct obd_device *obd, obd_count len, void *buf) if (rc) CERROR("Cannot process recover llog %d\n", rc); } + + if (rc == 0 && cld->cld_params != NULL) { + rc = mgc_process_log(obd, cld->cld_params); + if (rc == -ENOENT) { + CDEBUG(D_MGC, "There is no params " + "config file yet\n"); + rc = 0; + } + /* params log is optional */ + if (rc) + CERROR("%s: can't process params llog: rc = %d\n", + obd->obd_name, rc); + } config_log_put(cld); break; @@ -1955,7 +2202,7 @@ out: RETURN(rc); } -struct obd_ops mgc_obd_ops = { +static struct obd_ops mgc_obd_ops = { .o_owner = THIS_MODULE, .o_setup = mgc_setup, .o_precleanup = mgc_precleanup, @@ -1964,33 +2211,27 @@ struct obd_ops mgc_obd_ops = { .o_del_conn = client_import_del_conn, .o_connect = client_connect_import, .o_disconnect = client_disconnect_export, - //.o_enqueue = mgc_enqueue, - .o_cancel = mgc_cancel, - //.o_iocontrol = mgc_iocontrol, .o_set_info_async = mgc_set_info_async, .o_get_info = mgc_get_info, .o_import_event = mgc_import_event, - .o_llog_init = mgc_llog_init, - .o_llog_finish = mgc_llog_finish, .o_process_config = mgc_process_config, }; -int __init mgc_init(void) +static int __init mgc_init(void) { - return class_register_type(&mgc_obd_ops, NULL, NULL, - LUSTRE_MGC_NAME, NULL); + return class_register_type(&mgc_obd_ops, NULL, true, NULL, + LUSTRE_MGC_NAME, NULL); } -#ifdef __KERNEL__ -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); module_exit(mgc_exit); -#endif