X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fofd%2Fofd_dev.c;h=cb7a1b562891833b3516cc60770395add0cbec4e;hp=685bdf0a3e24e0bd597cdff65f90d79306eb3a93;hb=1711e26ae861c28829870c2433caf7ee232909cf;hpb=c25c4d60ac5fbede50a41da12f91de6e402b3c68 diff --git a/lustre/ofd/ofd_dev.c b/lustre/ofd/ofd_dev.c index 685bdf0..cb7a1b5 100644 --- a/lustre/ofd/ofd_dev.c +++ b/lustre/ofd/ofd_dev.c @@ -27,7 +27,6 @@ */ /* * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. * * lustre/ofd/ofd_dev.c * @@ -77,12 +76,12 @@ #include #include #include +#include #include "ofd_internal.h" /* Slab for OFD object allocation */ static struct kmem_cache *ofd_object_kmem; - static struct lu_kmem_descr ofd_caches[] = { { .ckd_cache = &ofd_object_kmem, @@ -162,7 +161,8 @@ out: * \retval negative value on error */ static int ofd_stack_init(const struct lu_env *env, - struct ofd_device *m, struct lustre_cfg *cfg) + struct ofd_device *m, struct lustre_cfg *cfg, + u32 *lmd_flags) { const char *dev = lustre_cfg_string(cfg, 0); struct lu_device *d; @@ -181,8 +181,13 @@ static int ofd_stack_init(const struct lu_env *env, } lmd = s2lsi(lmi->lmi_sb)->lsi_lmd; - if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK) - m->ofd_skip_lfsck = 1; + if (lmd) { + if (lmd->lmd_flags & LMD_FLG_SKIP_LFSCK) + m->ofd_skip_lfsck = 1; + if (lmd->lmd_flags & LMD_FLG_NO_PRECREATE) + m->ofd_no_precreate = 1; + *lmd_flags = lmd->lmd_flags; + } /* find bottom osd */ OBD_ALLOC(osdname, MTI_NAME_MAXLEN); @@ -251,11 +256,8 @@ static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m, } lu_site_purge(env, top->ld_site, ~0); - if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) { - LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_OTHER, NULL); - lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer); - } - + lu_site_print(env, top->ld_site, &top->ld_site->ls_obj_hash.nelems, + D_OTHER, lu_cdebug_printer); LASSERT(m->ofd_osd_exp); obd_disconnect(m->ofd_osd_exp); @@ -320,13 +322,7 @@ static bool match_symlink_param(char *param) sval = strchr(param, '='); if (sval != NULL) { paramlen = sval - param; - if (strncmp(param, "writethrough_cache_enable", - paramlen) == 0 || - strncmp(param, "readcache_max_filesize", - paramlen) == 0 || - strncmp(param, "read_cache_enable", - paramlen) == 0 || - strncmp(param, "brw_stats", paramlen) == 0) + if (strncmp(param, "brw_stats", paramlen) == 0) return true; } } @@ -359,11 +355,11 @@ static int ofd_process_config(const struct lu_env *env, struct lu_device *d, switch (cfg->lcfg_command) { case LCFG_PARAM: { - struct obd_device *obd = ofd_obd(m); /* For interoperability */ - struct cfg_interop_param *ptr = NULL; - struct lustre_cfg *old_cfg = NULL; - char *param = NULL; + struct cfg_interop_param *ptr = NULL; + struct lustre_cfg *old_cfg = NULL; + char *param = NULL; + ssize_t count; param = lustre_cfg_string(cfg, 1); if (param == NULL) { @@ -397,14 +393,16 @@ static int ofd_process_config(const struct lu_env *env, struct lu_device *d, break; } - rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg, - d->ld_obd); - if (rc > 0 || rc == -ENOSYS) { - CDEBUG(D_CONFIG, "pass param %s down the stack.\n", - param); - /* we don't understand; pass it on */ - rc = next->ld_ops->ldo_process_config(env, next, cfg); + count = class_modify_config(cfg, PARAM_OST, + &d->ld_obd->obd_kset.kobj); + if (count > 0) { + rc = 0; + break; } + CDEBUG(D_CONFIG, "pass param %s down the stack.\n", + param); + /* we don't understand; pass it on */ + rc = next->ld_ops->ldo_process_config(env, next, cfg); break; } case LCFG_SPTLRPC_CONF: { @@ -454,6 +452,14 @@ static int ofd_object_init(const struct lu_env *env, struct lu_object *o, RETURN(rc); } +static void ofd_object_free_rcu(struct rcu_head *head) +{ + struct ofd_object *of = container_of(head, struct ofd_object, + ofo_header.loh_rcu); + + kmem_cache_free(ofd_object_kmem, of); +} + /** * Implementation of lu_object_operations::loo_object_free. * @@ -475,7 +481,8 @@ static void ofd_object_free(const struct lu_env *env, struct lu_object *o) lu_object_fini(o); lu_object_header_fini(h); - OBD_SLAB_FREE_PTR(of, ofd_object_kmem); + OBD_FREE_PRE(of, sizeof(*of), "slab-freed"); + call_rcu(&of->ofo_header.loh_rcu, ofd_object_free_rcu); EXIT; } @@ -499,7 +506,7 @@ static int ofd_object_print(const struct lu_env *env, void *cookie, return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o); } -static struct lu_object_operations ofd_obj_ops = { +static const struct lu_object_operations ofd_obj_ops = { .loo_object_init = ofd_object_init, .loo_object_free = ofd_object_free, .loo_object_print = ofd_object_print @@ -537,6 +544,7 @@ static struct lu_object *ofd_object_alloc(const struct lu_env *env, lu_object_init(o, h, d); lu_object_add_top(h, o); o->lo_ops = &ofd_obj_ops; + range_lock_tree_init(&of->ofo_write_tree); RETURN(o); } else { RETURN(NULL); @@ -638,6 +646,8 @@ static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev, LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc); target_recovery_init(&ofd->ofd_lut, tgt_request_handle); + OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE, + (OBD_TIMEOUT_DEFAULT + 1) / 4); LASSERT(obd->obd_no_conn); spin_lock(&obd->obd_dev_lock); obd->obd_no_conn = 0; @@ -692,80 +702,14 @@ static int ofd_recovery_complete(const struct lu_env *env, /** * lu_device_operations matrix for OFD device. */ -static struct lu_device_operations ofd_lu_ops = { +static const struct lu_device_operations ofd_lu_ops = { .ldo_object_alloc = ofd_object_alloc, .ldo_process_config = ofd_process_config, .ldo_recovery_complete = ofd_recovery_complete, .ldo_prepare = ofd_prepare, }; -LPROC_SEQ_FOPS(lprocfs_nid_stats_clear); - -/** - * Initialize all needed procfs entries for OFD device. - * - * \param[in] ofd OFD device - * - * \retval 0 if successful - * \retval negative value on error - */ -static int ofd_procfs_init(struct ofd_device *ofd) -{ - struct obd_device *obd = ofd_obd(ofd); - struct proc_dir_entry *entry; - int rc = 0; - - ENTRY; - - /* lprocfs must be setup before the ofd so state can be safely added - * to /proc incrementally as the ofd is setup */ - obd->obd_vars = lprocfs_ofd_obd_vars; - rc = lprocfs_obd_setup(obd, false); - if (rc) { - CERROR("%s: lprocfs_obd_setup failed: %d.\n", - obd->obd_name, rc); - RETURN(rc); - } - - rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST); - if (rc) { - CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n", - obd->obd_name, rc); - GOTO(obd_cleanup, rc); - } - - entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL); - if (IS_ERR(entry)) { - rc = PTR_ERR(entry); - CERROR("%s: error %d setting up lprocfs for %s\n", - obd->obd_name, rc, "exports"); - GOTO(obd_cleanup, rc); - } - obd->obd_proc_exports_entry = entry; - - entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear", - obd, &lprocfs_nid_stats_clear_fops); - if (IS_ERR(entry)) { - rc = PTR_ERR(entry); - CERROR("%s: add proc entry 'clear' failed: %d.\n", - obd->obd_name, rc); - GOTO(obd_cleanup, rc); - } - - ofd_stats_counter_init(obd->obd_stats); - - rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST, - ofd_stats_counter_init); - if (rc) - GOTO(obd_cleanup, rc); - RETURN(0); -obd_cleanup: - lprocfs_obd_cleanup(obd); - lprocfs_free_obd_stats(obd); - - return rc; -} - +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0) /** * Expose OSD statistics to OFD layer. * @@ -779,8 +723,31 @@ obd_cleanup: */ static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd) { - struct obd_device *obd = ofd_obd(ofd); - struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd; + struct obd_device *obd = ofd_obd(ofd); + struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd; + struct kobj_type *osd_type; + int i; + + osd_type = get_ktype(&ofd->ofd_osd->dd_kobj); + for (i = 0; osd_type->default_attrs[i]; i++) { + if (strcmp(osd_type->default_attrs[i]->name, + "read_cache_enable") == 0) { + ofd->ofd_read_cache_enable = + osd_type->default_attrs[i]; + } + + if (strcmp(osd_type->default_attrs[i]->name, + "readcache_max_filesize") == 0) { + ofd->ofd_read_cache_max_filesize = + osd_type->default_attrs[i]; + } + + if (strcmp(osd_type->default_attrs[i]->name, + "writethrough_cache_enable") == 0) { + ofd->ofd_write_cache_enable = + osd_type->default_attrs[i]; + } + } if (obd->obd_proc_entry == NULL) return; @@ -788,21 +755,8 @@ static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd) lprocfs_add_symlink("brw_stats", obd->obd_proc_entry, "../../%s/%s/brw_stats", osd_obd->obd_type->typ_name, obd->obd_name); - - lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry, - "../../%s/%s/read_cache_enable", - osd_obd->obd_type->typ_name, obd->obd_name); - - lprocfs_add_symlink("readcache_max_filesize", - obd->obd_proc_entry, - "../../%s/%s/readcache_max_filesize", - osd_obd->obd_type->typ_name, obd->obd_name); - - lprocfs_add_symlink("writethrough_cache_enable", - obd->obd_proc_entry, - "../../%s/%s/writethrough_cache_enable", - osd_obd->obd_type->typ_name, obd->obd_name); } +#endif /** * Cleanup all procfs entries in OFD. @@ -813,6 +767,7 @@ static void ofd_procfs_fini(struct ofd_device *ofd) { struct obd_device *obd = ofd_obd(ofd); + tgt_tunables_fini(&ofd->ofd_lut); lprocfs_free_per_client_stats(obd); lprocfs_obd_cleanup(obd); lprocfs_free_obd_stats(obd); @@ -852,18 +807,19 @@ int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd) */ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) { - struct seq_server_site *ss = &ofd->ofd_seq_site; - struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev; - char *obd_name = ofd_name(ofd); - char *name = NULL; - int rc = 0; + struct seq_server_site *ss = &ofd->ofd_seq_site; + struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev; + char *obd_name = ofd_name(ofd); + char *name = NULL; + int len = strlen(obd_name) + 7; + int rc = 0; ss = &ofd->ofd_seq_site; lu->ld_site->ld_seq_site = ss; ss->ss_lu = lu->ld_site; ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index; - OBD_ALLOC(name, sizeof(obd_name) * 2 + 10); + OBD_ALLOC(name, len); if (name == NULL) return -ENOMEM; @@ -874,7 +830,7 @@ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name, LUSTRE_SEQ_SERVER, ss); if (rc) { - CERROR("%s : seq server init error %d\n", obd_name, rc); + CERROR("%s: seq server init error: rc = %d\n", obd_name, rc); GOTO(out_server, rc); } ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id; @@ -883,22 +839,13 @@ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) if (ss->ss_client_seq == NULL) GOTO(out_server, rc = -ENOMEM); - /* - * It always printed as "%p", so that the name is unique in the kernel, - * even if the filesystem is mounted twice. So sizeof(.) * 2 is enough. - */ - snprintf(name, sizeof(obd_name) * 2 + 7, "%p-super", obd_name); - rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA, - name, NULL); - if (rc) { - CERROR("%s : seq client init error %d\n", obd_name, rc); - GOTO(out_client, rc); - } + snprintf(name, len, "%s-super", obd_name); + seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA, + name, NULL); rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq); if (rc) { -out_client: seq_client_fini(ss->ss_client_seq); OBD_FREE_PTR(ss->ss_client_seq); ss->ss_client_seq = NULL; @@ -908,7 +855,7 @@ out_server: ss->ss_server_seq = NULL; } out_name: - OBD_FREE(name, sizeof(obd_name) * 2 + 10); + OBD_FREE(name, len); return rc; } @@ -930,6 +877,7 @@ static int ofd_set_info_hdl(struct tgt_session_info *tsi) void *key, *val = NULL; int keylen, vallen, rc = 0; bool is_grant_shrink; + ktime_t kstart = ktime_get(); ENTRY; @@ -962,6 +910,19 @@ static int ofd_set_info_hdl(struct tgt_session_info *tsi) if (is_grant_shrink) { body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY); + /* + * Because we already sync grant info with client when + * reconnect, grant info will be cleared for resent + * req, otherwise, outdated grant count in the rpc + * would de-sync grant counters + */ + if (lustre_msg_get_flags(req->rq_reqmsg) & + (MSG_RESENT | MSG_REPLAY)) { + DEBUG_REQ(D_CACHE, req, + "clear resent/replay req grant info"); + body->oa.o_valid &= ~OBD_MD_FLGRANT; + } + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); *repbody = *body; @@ -978,7 +939,7 @@ static int ofd_set_info_hdl(struct tgt_session_info *tsi) rc = -EOPNOTSUPP; } ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); RETURN(rc); } @@ -1021,7 +982,8 @@ int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd, } -static int ofd_lock_unlock_region(struct ldlm_namespace *ns, +static int ofd_lock_unlock_region(const struct lu_env *env, + struct ldlm_namespace *ns, struct ldlm_res_id *res_id, unsigned long long begin, unsigned long long end) @@ -1032,12 +994,12 @@ static int ofd_lock_unlock_region(struct ldlm_namespace *ns, LASSERT(begin <= end); - rc = tgt_extent_lock(ns, res_id, begin, end, &lh, LCK_PR, &flags); + rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags); if (rc != 0) return rc; CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh); - tgt_extent_unlock(&lh, LCK_PR); + tgt_data_unlock(&lh, LCK_PR); return 0; } @@ -1060,7 +1022,8 @@ static int ofd_lock_unlock_region(struct ldlm_namespace *ns, * \retval 0 if successful * \retval negative value on error */ -static int lock_zero_regions(struct ldlm_namespace *ns, +static int lock_zero_regions(const struct lu_env *env, + struct ldlm_namespace *ns, struct ldlm_res_id *res_id, struct fiemap *fiemap) { @@ -1076,7 +1039,7 @@ static int lock_zero_regions(struct ldlm_namespace *ns, if (fiemap_start[i].fe_logical > begin) { CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n", begin, fiemap_start[i].fe_logical); - rc = ofd_lock_unlock_region(ns, res_id, begin, + rc = ofd_lock_unlock_region(env, ns, res_id, begin, fiemap_start[i].fe_logical); if (rc) RETURN(rc); @@ -1088,7 +1051,7 @@ static int lock_zero_regions(struct ldlm_namespace *ns, if (begin < (fiemap->fm_start + fiemap->fm_length)) { CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n", begin, fiemap->fm_start + fiemap->fm_length); - rc = ofd_lock_unlock_region(ns, res_id, begin, + rc = ofd_lock_unlock_region(env, ns, res_id, begin, fiemap->fm_start + fiemap->fm_length); } @@ -1122,6 +1085,7 @@ static int ofd_get_info_hdl(struct tgt_session_info *tsi) void *key; int keylen; int replylen, rc = 0; + ktime_t kstart = ktime_get(); ENTRY; @@ -1189,7 +1153,7 @@ static int ofd_get_info_hdl(struct tgt_session_info *tsi) if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS && fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) { ost_fid_build_resid(fid, &fti->fti_resid); - rc = lock_zero_regions(ofd->ofd_namespace, + rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace, &fti->fti_resid, fiemap); if (rc == 0) rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, @@ -1236,7 +1200,7 @@ out_put: rc = -EOPNOTSUPP; } ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); RETURN(rc); } @@ -1261,6 +1225,7 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) struct ofd_object *fo; __u64 flags = 0; enum ldlm_mode lock_mode = LCK_PR; + ktime_t kstart = ktime_get(); bool srvlock; int rc; ENTRY; @@ -1281,7 +1246,8 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH)) lock_mode = LCK_PW; - rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace, + rc = tgt_extent_lock(tsi->tsi_env, + tsi->tsi_tgt->lut_obd->obd_namespace, &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh, lock_mode, &flags); if (rc != 0) @@ -1321,10 +1287,10 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) ofd_object_put(tsi->tsi_env, fo); out: if (srvlock) - tgt_extent_unlock(&lh, lock_mode); + tgt_data_unlock(&lh, lock_mode); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); repbody->oa.o_valid |= OBD_MD_FLFLAGS; repbody->oa.o_flags = OBD_FL_FLUSH; @@ -1351,6 +1317,7 @@ static int ofd_setattr_hdl(struct tgt_session_info *tsi) struct ost_body *repbody; struct ldlm_resource *res; struct ofd_object *fo; + ktime_t kstart = ktime_get(); int rc = 0; ENTRY; @@ -1399,7 +1366,7 @@ static int ofd_setattr_hdl(struct tgt_session_info *tsi) OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); EXIT; out_put: ofd_object_put(tsi->tsi_env, fo); @@ -1536,16 +1503,20 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) u64 seq = ostid_seq(&oa->o_oi); u64 oid = ostid_id(&oa->o_oi); struct ofd_seq *oseq; - s64 diff; - int rc = 0; int sync_trans = 0; long granted = 0; + ktime_t kstart = ktime_get(); + s64 diff; + int rc = 0; ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS)) RETURN(-EROFS); + if (ofd->ofd_no_precreate) + return -EPERM; + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); if (repbody == NULL) RETURN(-ENOMEM); @@ -1665,7 +1636,7 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) GOTO(out, rc = -EINVAL); } - if (diff < 0) { + if (diff <= -OST_MAX_PRECREATE) { /* LU-5648 */ CERROR("%s: invalid precreate request for " DOSTID", last_id %llu. " @@ -1673,6 +1644,15 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) ofd_name(ofd), POSTID(&oa->o_oi), ofd_seq_last_oid(oseq)); GOTO(out, rc = -EINVAL); + } else if (diff < 0) { + LCONSOLE(D_INFO, + "%s: MDS LAST_ID "DFID" (%llu) is %lld behind OST LAST_ID "DFID" (%llu), trust the OST\n", + ofd_name(ofd), PFID(&oa->o_oi.oi_fid), + oid, -diff, PFID(&oseq->os_oi.oi_fid), + ofd_seq_last_oid(oseq)); + /* Let MDS know that we are so far ahead. */ + rc = ostid_set_id(&rep_oa->o_oi, + ofd_seq_last_oid(oseq) + 1); } } } @@ -1681,6 +1661,7 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) u64 next_id; int created = 0; int count; + int rc2; if (!(oa->o_valid & OBD_MD_FLFLAGS) || !(oa->o_flags & OBD_FL_DELORPHAN)) { @@ -1704,18 +1685,12 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) * (possibly filling the OST), only precreate the last batch. * LFSCK will eventually clean up any orphans. LU-14 */ if (diff > 5 * OST_MAX_PRECREATE) { + /* Message below is checked in conf-sanity test_122b */ + LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %lld higher than LAST_ID "DOSTID", only precreating the last %u objects. OST replaced or reformatted?\n", + ofd_name(ofd), POSTID(&oa->o_oi), diff, + POSTID(&oseq->os_oi), + OST_MAX_PRECREATE / 2); diff = OST_MAX_PRECREATE / 2; - LCONSOLE_WARN("%s: Too many FIDs to precreate " - "OST replaced or reformatted: " - "LFSCK will clean up", - ofd_name(ofd)); - - CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over " - "%u larger than the LAST_ID "DOSTID", only " - "precreating the last %lld objects.\n", - ofd_name(ofd), POSTID(&oa->o_oi), - 5 * OST_MAX_PRECREATE, - POSTID(&oseq->os_oi), diff); ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff); } @@ -1768,26 +1743,19 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) granted = 0; } - rc = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq)); + rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq)); + rc = rc ? : rc2; } EXIT; ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); + if (unlikely(!oseq->os_last_id_synced)) + oseq->os_last_id_synced = 1; out: mutex_unlock(&oseq->os_create_lock); out_nolock: - if (rc == 0) { -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0) - struct ofd_thread_info *info = ofd_info(tsi->tsi_env); - struct lu_fid *fid = &info->fti_fid; - - /* For compatible purpose, it needs to convert back to - * OST ID before put it on wire. */ - *fid = rep_oa->o_oi.oi_fid; - fid_to_ostid(fid, &rep_oa->o_oi); -#endif + if (rc == 0) rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP; - } ofd_seq_put(tsi->tsi_env, oseq); out_sem: @@ -1813,6 +1781,7 @@ static int ofd_destroy_hdl(struct tgt_session_info *tsi) struct ofd_device *ofd = ofd_exp(tsi->tsi_exp); struct ofd_thread_info *fti = tsi2ofd_info(tsi); struct lu_fid *fid = &fti->fti_fid; + ktime_t kstart = ktime_get(); u64 oid; u32 count; int rc = 0; @@ -1874,7 +1843,7 @@ static int ofd_destroy_hdl(struct tgt_session_info *tsi) } ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); GOTO(out, rc); @@ -1896,6 +1865,7 @@ out: */ static int ofd_statfs_hdl(struct tgt_session_info *tsi) { + ktime_t kstart = ktime_get(); struct obd_statfs *osfs; int rc; @@ -1915,7 +1885,7 @@ static int ofd_statfs_hdl(struct tgt_session_info *tsi) rc = -EINPROGRESS; ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); RETURN(rc); } @@ -1938,6 +1908,7 @@ static int ofd_sync_hdl(struct tgt_session_info *tsi) struct ofd_thread_info *fti = tsi2ofd_info(tsi); struct ofd_device *ofd = ofd_exp(tsi->tsi_exp); struct ofd_object *fo = NULL; + ktime_t kstart = ktime_get(); int rc = 0; ENTRY; @@ -1958,7 +1929,7 @@ static int ofd_sync_hdl(struct tgt_session_info *tsi) GOTO(put, rc); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); if (fo == NULL) RETURN(0); @@ -1980,6 +1951,129 @@ put: } /** + * OFD request handler for OST_FALLOCATE RPC. + * + * This is part of request processing. Validate request fields, + * preallocate the given OFD object and pack reply. + * + * \param[in] tsi target session environment for this request + * + * \retval 0 if successful + * \retval negative value on error + */ +static int ofd_fallocate_hdl(struct tgt_session_info *tsi) +{ + struct obdo *oa = &tsi->tsi_ost_body->oa; + struct ost_body *repbody; + struct ofd_thread_info *info = tsi2ofd_info(tsi); + struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace; + struct ldlm_resource *res; + struct ofd_object *fo; + __u64 flags = 0; + __u64 valid; + struct lustre_handle lh = { 0, }; + int rc, mode; + __u64 start, end; + bool srvlock; + ktime_t kstart = ktime_get(); + + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); + if (repbody == NULL) + RETURN(err_serious(-ENOMEM)); + + /* + * fallocate start and end are passed in o_size, o_blocks + * on the wire. + */ + if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) != + (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) + RETURN(err_serious(-EPROTO)); + + start = oa->o_size; + end = oa->o_blocks; + mode = oa->o_falloc_mode; + /* + * mode == 0 (which is standard prealloc) and PUNCH is supported + * Rest of mode options are not supported yet. + */ + if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) + RETURN(-EOPNOTSUPP); + + /* PUNCH_HOLE mode should always be accompanied with KEEP_SIZE flag + * Check that and add the missing flag for such invalid call with + * warning. + */ + if (mode & FALLOC_FL_PUNCH_HOLE && !(mode & FALLOC_FL_KEEP_SIZE)) { + CWARN("%s: PUNCH mode misses KEEP_SIZE flag, setting it\n", + tsi->tsi_tgt->lut_obd->obd_name); + mode |= FALLOC_FL_KEEP_SIZE; + } + + repbody->oa.o_oi = oa->o_oi; + repbody->oa.o_valid = OBD_MD_FLID; + + srvlock = oa->o_valid & OBD_MD_FLFLAGS && + oa->o_flags & OBD_FL_SRVLOCK; + + if (srvlock) { + rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, + start, end, &lh, LCK_PW, &flags); + if (rc != 0) + RETURN(rc); + } + + fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp), + &tsi->tsi_fid); + if (IS_ERR(fo)) + GOTO(out, rc = PTR_ERR(fo)); + + valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID | + OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME; + la_from_obdo(&info->fti_attr, oa, valid); + + rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode, + &info->fti_attr, oa); + if (rc) + GOTO(out_put, rc); + + rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr); + if (rc == 0) + obdo_from_la(&repbody->oa, &info->fti_attr, OFD_VALID_FLAGS); + else + rc = 0; + + ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC, + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); + + EXIT; +out_put: + ofd_object_put(tsi->tsi_env, fo); +out: + if (srvlock) + tgt_data_unlock(&lh, LCK_PW); + if (rc == 0) { + res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid, + LDLM_EXTENT, 0); + if (!IS_ERR(res)) { + struct ost_lvb *res_lvb; + + ldlm_res_lvbo_update(res, NULL, 0); + res_lvb = res->lr_lvb_data; + /* Blocks */ + repbody->oa.o_valid |= OBD_MD_FLBLOCKS; + repbody->oa.o_blocks = res_lvb->lvb_blocks; + /* Size */ + repbody->oa.o_valid |= OBD_MD_FLSIZE; + repbody->oa.o_size = res_lvb->lvb_size; + + ldlm_resource_putref(res); + } + } + + RETURN(rc); +} + +/** * OFD request handler for OST_PUNCH RPC. * * This is part of request processing. Validate request fields, @@ -2000,17 +2094,15 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi) struct ofd_object *fo; __u64 flags = 0; struct lustre_handle lh = { 0, }; - int rc; __u64 start, end; bool srvlock; + ktime_t kstart = ktime_get(); + int rc; ENTRY; OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val); - /* check that we do support OBD_CONNECT_TRUNCLOCK. */ - CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK); - if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) != (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) RETURN(err_serious(-EPROTO)); @@ -2038,8 +2130,8 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi) oa->o_flags & OBD_FL_SRVLOCK; if (srvlock) { - rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh, - LCK_PW, &flags); + rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start, + end, &lh, LCK_PW, &flags); if (rc != 0) RETURN(rc); } @@ -2064,13 +2156,13 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi) GOTO(out_put, rc); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); EXIT; out_put: ofd_object_put(tsi->tsi_env, fo); out: if (srvlock) - tgt_extent_unlock(&lh, LCK_PW); + tgt_data_unlock(&lh, LCK_PW); if (rc == 0) { /* we do not call this before to avoid lu_object_find() in * ->lvbo_update() holding another reference on the object. @@ -2129,7 +2221,8 @@ static int ofd_ladvise_prefetch(const struct lu_env *env, PTLRPC_MAX_BRW_PAGES; rnb.rnb_offset = start_index << PAGE_SHIFT; rnb.rnb_len = nr_local << PAGE_SHIFT; - rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, dbt); + rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, + PTLRPC_MAX_BRW_PAGES, dbt); if (unlikely(rc < 0)) break; nr_local = rc; @@ -2249,7 +2342,7 @@ static int ofd_ladvise_hdl(struct tgt_session_info *tsi) ioo.ioo_oid = body->oa.o_oi; ioo.ioo_bufcnt = 1; - rc = tgt_extent_lock(exp->exp_obd->obd_namespace, + rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace, &tsi->tsi_resid, start, end - 1, &lockh, LCK_PR, &flags); if (rc != 0) @@ -2258,7 +2351,7 @@ static int ofd_ladvise_hdl(struct tgt_session_info *tsi) req->rq_status = ofd_ladvise_prefetch(env, fo, tbc->local, start, end, dbt); - tgt_extent_unlock(&lockh, LCK_PR); + tgt_data_unlock(&lockh, LCK_PR); break; case LU_LADVISE_DONTNEED: rc = dt_ladvise(env, dob, ladvise->lla_start, @@ -2289,6 +2382,7 @@ static int ofd_quotactl(struct tgt_session_info *tsi) { struct obd_quotactl *oqctl, *repoqc; struct lu_nodemap *nodemap; + ktime_t kstart = ktime_get(); int id; int rc; @@ -2317,6 +2411,10 @@ static int ofd_quotactl(struct tgt_session_info *tsi) id = nodemap_map_id(nodemap, NODEMAP_GID, NODEMAP_CLIENT_TO_FS, repoqc->qc_id); + else if (oqctl->qc_type == PRJQUOTA) + id = nodemap_map_id(nodemap, NODEMAP_PROJID, + NODEMAP_CLIENT_TO_FS, + repoqc->qc_id); nodemap_putref(nodemap); @@ -2326,7 +2424,7 @@ static int ofd_quotactl(struct tgt_session_info *tsi) rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); if (repoqc->qc_id != id) swap(repoqc->qc_id, id); @@ -2335,29 +2433,6 @@ static int ofd_quotactl(struct tgt_session_info *tsi) } /** - * Calculate the amount of time for lock prolongation. - * - * This is helper for ofd_prolong_extent_locks() function to get - * the timeout extra time. - * - * \param[in] req current request - * - * \retval amount of time to extend the timeout with - */ -static inline time64_t prolong_timeout(struct ptlrpc_request *req) -{ - struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt; - time64_t req_timeout; - - if (AT_OFF) - return obd_timeout / 2; - - req_timeout = req->rq_deadline - req->rq_arrival_time.tv_sec; - return max_t(time64_t, at_est2timeout(at_get(&svcpt->scp_at_estimate)), - req_timeout); -} - -/** * Prolong lock timeout for the given extent. * * This function finds all locks related with incoming request and @@ -2533,14 +2608,14 @@ static int ofd_rw_hpreq_check(struct ptlrpc_request *req) rnb += ioo->ioo_bufcnt - 1; pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1; - DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID - " (%llu->%llu)\n", tgt_name(tsi->tsi_tgt), - current->comm, PFID(&tsi->tsi_fid), pa.lpa_extent.start, - pa.lpa_extent.end); + DEBUG_REQ(D_RPCTRACE, req, + "%s %s: refresh rw locks for "DFID" (%llu->%llu)", + tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid), + pa.lpa_extent.start, pa.lpa_extent.end); ofd_prolong_extent_locks(tsi, &pa); - CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n", + CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n", tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req); if (pa.lpa_blocks_cnt > 0) @@ -2765,26 +2840,27 @@ TGT_RPC_HANDLER(OST_FIRST_OPC, 0, OST_SET_INFO, ofd_set_info_hdl, &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION), TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl), -TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl), -TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_SETATTR, ofd_setattr_hdl), -TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR, +TGT_OST_HDL(HAS_REPLY | IS_MUTABLE, OST_CREATE, ofd_create_hdl), -TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR, +TGT_OST_HDL(HAS_REPLY | IS_MUTABLE, OST_DESTROY, ofd_destroy_hdl), -TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl), -TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO, - OST_BRW_READ, tgt_brw_read, +TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl), +TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read, ofd_hp_brw), /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */ -TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write, +TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write, ofd_hp_brw), -TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR, +TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_PUNCH, ofd_punch_hdl, ofd_hp_punch), -TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl), -TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl), -TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl), +TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek), }; static struct tgt_opc_slice ofd_common_slice[] = { @@ -2889,6 +2965,7 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, struct lu_fid fid; struct nm_config_file *nodemap_config; struct obd_device_target *obt; + u32 lmd_flags = 0; int rc; ENTRY; @@ -2906,21 +2983,18 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, obt = &obd->u.obt; obt->obt_magic = OBT_MAGIC; - m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT; - m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT; - spin_lock_init(&m->ofd_flags_lock); m->ofd_raid_degraded = 0; - m->ofd_checksum_t10pi_enforce = 0; - m->ofd_syncjournal = 0; + m->ofd_sync_journal = 0; ofd_slc_set(m); m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT; m->ofd_seq_count = 0; - init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq); INIT_LIST_HEAD(&m->ofd_inconsistency_list); spin_lock_init(&m->ofd_inconsistency_lock); + m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */ + spin_lock_init(&m->ofd_batch_lock); init_rwsem(&m->ofd_lastid_rwsem); @@ -2929,12 +3003,6 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, /* set this lu_device to obd, because error handling need it */ obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev; - rc = ofd_procfs_init(m); - if (rc) { - CERROR("Can't init ofd lprocfs, rc %d\n", rc); - RETURN(rc); - } - /* No connection accepted until configurations will finish */ spin_lock(&obd->obd_dev_lock); obd->obd_no_conn = 1; @@ -2951,15 +3019,18 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, info = ofd_info_init(env, NULL); if (info == NULL) - GOTO(err_fini_proc, rc = -EFAULT); + RETURN(-EFAULT); - rc = ofd_stack_init(env, m, cfg); + rc = ofd_stack_init(env, m, cfg, &lmd_flags); if (rc) { - CERROR("Can't init device stack, rc %d\n", rc); - GOTO(err_fini_proc, rc); + CERROR("%s: can't init device stack, rc %d\n", + obd->obd_name, rc); + RETURN(rc); } +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0) ofd_procfs_add_brw_stats_symlink(m); +#endif snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s", "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid); @@ -2967,8 +3038,13 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, LDLM_NAMESPACE_SERVER, LDLM_NAMESPACE_GREEDY, LDLM_NS_TYPE_OST); - if (m->ofd_namespace == NULL) - GOTO(err_fini_stack, rc = -ENOMEM); + if (IS_ERR(m->ofd_namespace)) { + rc = PTR_ERR(m->ofd_namespace); + CERROR("%s: unable to create server namespace: rc = %d\n", + obd->obd_name, rc); + m->ofd_namespace = NULL; + GOTO(err_fini_stack, rc); + } /* set obd_namespace for compatibility with old code */ obd->obd_namespace = m->ofd_namespace; ldlm_register_intent(m->ofd_namespace, ofd_intent_policy); @@ -2984,19 +3060,27 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, if (rc) GOTO(err_free_ns, rc); + if (lmd_flags & LMD_FLG_SKIP_LFSCK) + m->ofd_skip_lfsck = 1; + if (lmd_flags & LMD_FLG_LOCAL_RECOV) + m->ofd_lut.lut_local_recovery = 1; + + rc = ofd_tunables_init(m); + if (rc) + GOTO(err_fini_lut, rc); + tgd->tgd_reserved_pcnt = 0; m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size; - m->ofd_cksum_types_supported = - obd_cksum_types_supported_server(obd->obd_name); m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT; if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks < OFD_PRECREATE_SMALL_FS) m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL; + m->ofd_atime_diff = OFD_DEF_ATIME_DIFF; rc = ofd_fs_setup(env, m, obd); if (rc) - GOTO(err_fini_lut, rc); + GOTO(err_fini_proc, rc); fid.f_seq = FID_SEQ_LOCAL_NAME; fid.f_oid = 1; @@ -3032,6 +3116,8 @@ err_fini_los: m->ofd_los = NULL; err_fini_fs: ofd_fs_cleanup(env, m); +err_fini_proc: + ofd_procfs_fini(m); err_fini_lut: tgt_fini(env, &m->ofd_lut); err_free_ns: @@ -3039,8 +3125,6 @@ err_free_ns: obd->obd_namespace = m->ofd_namespace = NULL; err_fini_stack: ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev); -err_fini_proc: - ofd_procfs_fini(m); return rc; } @@ -3071,6 +3155,7 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m) obd_exports_barrier(obd); obd_zombie_barrier(); + ofd_procfs_fini(m); tgt_fini(env, &m->ofd_lut); ofd_stop_inconsistency_verification_thread(m); lfsck_degister(env, m->ofd_osd); @@ -3083,8 +3168,11 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m) d->ld_obd->obd_namespace = m->ofd_namespace = NULL; } + ofd_access_log_delete(m->ofd_access_log); + m->ofd_access_log = NULL; + ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev); - ofd_procfs_fini(m); + LASSERT(atomic_read(&d->ld_ref) == 0); server_put_mount(obd->obd_name, true); EXIT; @@ -3168,7 +3256,7 @@ static struct lu_device *ofd_device_alloc(const struct lu_env *env, /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */ LU_TYPE_INIT_FINI(ofd, &ofd_thread_key); -static struct lu_device_type_operations ofd_device_type_ops = { +static const struct lu_device_type_operations ofd_device_type_ops = { .ldto_init = ofd_type_init, .ldto_fini = ofd_type_fini, @@ -3198,20 +3286,28 @@ static struct lu_device_type ofd_device_type = { */ static int __init ofd_init(void) { - int rc; + int rc; rc = lu_kmem_init(ofd_caches); if (rc) return rc; - rc = ofd_fmd_init(); - if (rc) { - lu_kmem_fini(ofd_caches); - return(rc); - } + rc = ofd_access_log_module_init(); + if (rc) + goto out_caches; - rc = class_register_type(&ofd_obd_ops, NULL, true, NULL, + rc = class_register_type(&ofd_obd_ops, NULL, true, LUSTRE_OST_NAME, &ofd_device_type); + if (rc) + goto out_ofd_access_log; + + return 0; + +out_ofd_access_log: + ofd_access_log_module_exit(); +out_caches: + lu_kmem_fini(ofd_caches); + return rc; } @@ -3223,9 +3319,9 @@ static int __init ofd_init(void) */ static void __exit ofd_exit(void) { - ofd_fmd_exit(); - lu_kmem_fini(ofd_caches); class_unregister_type(LUSTRE_OST_NAME); + ofd_access_log_module_exit(); + lu_kmem_fini(ofd_caches); } MODULE_AUTHOR("OpenSFS, Inc. ");