X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fobdfilter%2Ffilter.c;h=c8d25cf9e4c252de908b37436fee2843116ac0c4;hb=8e417baed54e23cfa35e3caff9b46144e083f1a8;hp=0e10ec8a0b80ee4cd7db6df435e47e017b43d7ea;hpb=9d3ad9fba3e244735f0994c3910e9de00e2a7f4d;p=fs%2Flustre-release.git diff --git a/lustre/obdfilter/filter.c b/lustre/obdfilter/filter.c index 0e10ec8..c8d25cf 100644 --- a/lustre/obdfilter/filter.c +++ b/lustre/obdfilter/filter.c @@ -80,19 +80,48 @@ #include "filter_internal.h" /* Group 0 is no longer a legal group, to catch uninitialized IDs */ -#define FILTER_MIN_GROUPS FILTER_GROUP_MDS0 +#define FILTER_MIN_GROUPS FILTER_GROUP_MDS1_N_BASE static struct lvfs_callback_ops filter_lvfs_ops; cfs_mem_cache_t *ll_fmd_cachep; static void filter_commit_cb(struct obd_device *obd, __u64 transno, void *cb_data, int error) { - obd_transno_commit_cb(obd, transno, error); + struct obd_export *exp = cb_data; + LASSERT(exp->exp_obd == obd); + obd_transno_commit_cb(obd, transno, exp, error); + atomic_dec(&exp->exp_cb_count); + class_export_put(exp); +} + +int filter_version_get_check(struct obd_export *exp, + struct obd_trans_info *oti, struct inode *inode) +{ + __u64 curr_version; + + if (inode == NULL || oti == NULL) + RETURN(0); + + curr_version = fsfilt_get_version(exp->exp_obd, inode); + if ((__s64)curr_version == -EOPNOTSUPP) + RETURN(0); + /* VBR: version is checked always because costs nothing */ + if (oti->oti_pre_version != 0 && + oti->oti_pre_version != curr_version) { + CDEBUG(D_INODE, "Version mismatch "LPX64" != "LPX64"\n", + oti->oti_pre_version, curr_version); + spin_lock(&exp->exp_lock); + exp->exp_vbr_failed = 1; + spin_unlock(&exp->exp_lock); + RETURN (-EOVERFLOW); + } + oti->oti_pre_version = curr_version; + RETURN(0); } /* Assumes caller has already pushed us into the kernel context. */ -int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti, - int rc, int force_sync) +int filter_finish_transno(struct obd_export *exp, struct inode *inode, + struct obd_trans_info *oti, int rc, int force_sync) { struct filter_obd *filter = &exp->exp_obd->u.filter; struct filter_export_data *fed = &exp->exp_filter_data; @@ -109,24 +138,28 @@ int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti, RETURN(rc); /* we don't allocate new transnos for replayed requests */ + spin_lock(&filter->fo_translock); if (oti->oti_transno == 0) { - spin_lock(&filter->fo_translock); last_rcvd = le64_to_cpu(filter->fo_fsd->lsd_last_transno) + 1; filter->fo_fsd->lsd_last_transno = cpu_to_le64(last_rcvd); - spin_unlock(&filter->fo_translock); - oti->oti_transno = last_rcvd; } else { - spin_lock(&filter->fo_translock); last_rcvd = oti->oti_transno; if (last_rcvd > le64_to_cpu(filter->fo_fsd->lsd_last_transno)) filter->fo_fsd->lsd_last_transno = cpu_to_le64(last_rcvd); + } + oti->oti_transno = last_rcvd; + if (last_rcvd <= le64_to_cpu(lcd->lcd_last_transno)) { spin_unlock(&filter->fo_translock); + LBUG(); } lcd->lcd_last_transno = cpu_to_le64(last_rcvd); + lcd->lcd_pre_versions[0] = cpu_to_le64(oti->oti_pre_version); + lcd->lcd_last_xid = cpu_to_le64(oti->oti_xid); + spin_unlock(&filter->fo_translock); - /* could get xid from oti, if it's ever needed */ - lcd->lcd_last_xid = 0; + if (inode) + fsfilt_set_version(exp->exp_obd, inode, last_rcvd); off = fed->fed_lr_off; if (off <= 0) { @@ -134,18 +167,20 @@ int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti, fed->fed_lr_idx, fed->fed_lr_off); err = -EINVAL; } else { + class_export_get(exp); /* released when the cb is called */ + atomic_inc(&exp->exp_cb_count); if (!force_sync) force_sync = fsfilt_add_journal_cb(exp->exp_obd, last_rcvd, oti->oti_handle, filter_commit_cb, - NULL); + exp); err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, lcd, sizeof(*lcd), &off, force_sync | exp->exp_need_sync); if (force_sync) - filter_commit_cb(exp->exp_obd, last_rcvd, NULL, err); + filter_commit_cb(exp->exp_obd, last_rcvd, exp, err); } if (err) { log_pri = D_ERROR; @@ -183,7 +218,7 @@ static int lprocfs_init_rw_stats(struct obd_device *obd, num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) + LPROC_FILTER_LAST - 1; - *stats = lprocfs_alloc_stats(num_stats, 0); + *stats = lprocfs_alloc_stats(num_stats, LPROCFS_STATS_FLAG_NOPERCPU); if (*stats == NULL) return -ENOMEM; @@ -202,12 +237,9 @@ static int filter_export_stats_init(struct obd_device *obd, struct obd_export *exp, void *client_nid) { - struct filter_export_data *fed = &exp->exp_filter_data; int rc, newnid = 0; ENTRY; - init_brw_stats(&fed->fed_brw_stats); - if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid)) /* Self-export gets no proc entry */ RETURN(0); @@ -245,8 +277,9 @@ static int filter_export_stats_init(struct obd_device *obd, if (rc) RETURN(rc); /* Always add in ldlm_stats */ - tmp->nid_ldlm_stats = lprocfs_alloc_stats(LDLM_LAST_OPC - - LDLM_FIRST_OPC, 0); + tmp->nid_ldlm_stats = + lprocfs_alloc_stats(LDLM_LAST_OPC - LDLM_FIRST_OPC, + LPROCFS_STATS_FLAG_NOPERCPU); if (tmp->nid_ldlm_stats == NULL) return -ENOMEM; @@ -331,6 +364,9 @@ static int filter_client_add(struct obd_device *obd, struct obd_export *exp, rc = PTR_ERR(handle); CERROR("unable to start transaction: rc %d\n", rc); } else { + fed->fed_lcd->lcd_last_epoch = + filter->fo_fsd->lsd_start_epoch; + exp->exp_last_request_time = cfs_time_current_sec(); rc = fsfilt_add_journal_cb(obd, 0, handle, target_client_add_cb, exp); if (rc == 0) { @@ -357,12 +393,13 @@ static int filter_client_add(struct obd_device *obd, struct obd_export *exp, RETURN(0); } +struct lsd_client_data zero_lcd; /* globals are implicitly zeroed */ + static int filter_client_free(struct obd_export *exp) { struct filter_export_data *fed = &exp->exp_filter_data; struct filter_obd *filter = &exp->exp_obd->u.filter; struct obd_device *obd = exp->exp_obd; - struct lsd_client_data zero_lcd; struct lvfs_run_ctxt saved; int rc; loff_t off; @@ -375,13 +412,13 @@ static int filter_client_free(struct obd_export *exp) if (strcmp(fed->fed_lcd->lcd_uuid, obd->obd_uuid.uuid ) == 0) GOTO(free, 0); - CDEBUG(D_INFO, "freeing client at idx %u, offset %lld with UUID '%s'\n", - fed->fed_lr_idx, fed->fed_lr_off, fed->fed_lcd->lcd_uuid); - LASSERT(filter->fo_last_rcvd_slots != NULL); off = fed->fed_lr_off; + CDEBUG(D_INFO, "freeing client at idx %u, offset %lld with UUID '%s'\n", + fed->fed_lr_idx, fed->fed_lr_off, fed->fed_lcd->lcd_uuid); + /* Don't clear fed_lr_idx here as it is likely also unset. At worst * we leak a client slot that will be cleaned on the next recovery. */ if (off <= 0) { @@ -399,23 +436,26 @@ static int filter_client_free(struct obd_export *exp) } if (!(exp->exp_flags & OBD_OPT_FAILOVER)) { - memset(&zero_lcd, 0, sizeof zero_lcd); + /* Don't force sync on disconnect if aborting recovery, + * or it does num_clients * num_osts. b=17194 */ + int need_sync = exp->exp_need_sync && + !(exp->exp_flags&OBD_OPT_ABORT_RECOV); push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_lcd, - sizeof(zero_lcd), &off, - (!exp->exp_libclient || - exp->exp_need_sync)); + sizeof(zero_lcd), &off, 0); + + /* Make sure the server's last_transno is up to date. Do this + * after the client is freed so we know all the client's + * transactions have been committed. */ if (rc == 0) - /* update server's transno */ filter_update_server_data(obd, filter->fo_rcvd_filp, - filter->fo_fsd, - !exp->exp_libclient); + filter->fo_fsd, need_sync); pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); CDEBUG(rc == 0 ? D_INFO : D_ERROR, - "zeroing out client %s at idx %u (%llu) in %s rc %d\n", + "zero out client %s at idx %u/%llu in %s %ssync rc %d\n", fed->fed_lcd->lcd_uuid, fed->fed_lr_idx, fed->fed_lr_off, - LAST_RCVD, rc); + LAST_RCVD, need_sync ? "" : "a", rc); } if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) { @@ -541,7 +581,7 @@ struct filter_mod_data *filter_fmd_get(struct obd_export *exp, struct filter_export_data *fed = &exp->exp_filter_data; struct filter_mod_data *found = NULL, *fmd_new = NULL; - OBD_SLAB_ALLOC(fmd_new, ll_fmd_cachep, CFS_ALLOC_IO, sizeof(*fmd_new)); + OBD_SLAB_ALLOC_PTR_GFP(fmd_new, ll_fmd_cachep, CFS_ALLOC_IO); spin_lock(&fed->fed_lock); found = filter_fmd_find_nolock(&exp->exp_obd->u.filter,fed,objid,group); @@ -617,7 +657,7 @@ static int filter_init_export(struct obd_export *exp) static int filter_free_server_data(struct filter_obd *filter) { - OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd)); + OBD_FREE_PTR(filter->fo_fsd); filter->fo_fsd = NULL; OBD_FREE(filter->fo_last_rcvd_slots, LR_MAX_CLIENTS / 8); filter->fo_last_rcvd_slots = NULL; @@ -638,7 +678,6 @@ int filter_update_server_data(struct obd_device *obd, struct file *filp, CDEBUG(D_INODE, "server last_mount: "LPU64"\n", le64_to_cpu(fsd->lsd_mount_count)); - fsd->lsd_compat14 = fsd->lsd_last_transno; rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off, force_sync); if (rc) CERROR("error writing lr_server_data: rc = %d\n", rc); @@ -682,6 +721,7 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp) struct inode *inode = filp->f_dentry->d_inode; unsigned long last_rcvd_size = i_size_read(inode); __u64 mount_count; + __u32 start_epoch; int cl_idx; loff_t off = 0; int rc; @@ -753,7 +793,11 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp) GOTO(err_fsd, rc = -EINVAL); } - CDEBUG(D_INODE, "%s: server last_transno : "LPU64"\n", + start_epoch = le32_to_cpu(fsd->lsd_start_epoch); + + CDEBUG(D_INODE, "%s: server start_epoch : %#x\n", + obd->obd_name, start_epoch); + CDEBUG(D_INODE, "%s: server last_transno : "LPX64"\n", obd->obd_name, le64_to_cpu(fsd->lsd_last_transno)); CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n", obd->obd_name, mount_count + 1); @@ -833,12 +877,16 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp) /* can't fail for existing client */ LASSERTF(rc == 0, "rc = %d\n", rc); - lcd = NULL; + /* VBR: set export last committed */ + exp->exp_last_committed = last_rcvd; spin_lock(&exp->exp_lock); exp->exp_connecting = 0; exp->exp_in_recovery = 0; spin_unlock(&exp->exp_lock); + spin_lock_bh(&obd->obd_processing_task_lock); obd->obd_max_recoverable_clients++; + spin_unlock_bh(&obd->obd_processing_task_lock); + lcd = NULL; class_export_put(exp); } @@ -855,7 +903,7 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp) obd->obd_last_committed = le64_to_cpu(fsd->lsd_last_transno); - target_recovery_init(obd, ost_handle); + target_recovery_init(&filter->fo_lut, ost_handle); out: filter->fo_mount_count = mount_count + 1; @@ -955,7 +1003,9 @@ static int filter_update_last_group(struct obd_device *obd, int group) CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc); GOTO(cleanup, rc); } - LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS); + LASSERTF(off == 0 || CHECK_MDS_GROUP(last_group), + "off = %llu and last_group = %d\n", off, last_group); + CDEBUG(D_INODE, "%s: previous %d, new %d\n", obd->obd_name, last_group, group); @@ -1145,8 +1195,6 @@ static int filter_read_groups(struct obd_device *obd, int last_group, down(&filter->fo_init_lock); old_count = filter->fo_group_count; for (group = old_count; group <= last_group; group++) { - if (group == 0) - continue; /* no group zero */ rc = filter_read_group_internal(obd, group, create); if (rc != 0) @@ -1245,7 +1293,7 @@ static int filter_prep_groups(struct obd_device *obd) if (off == 0) { last_group = FILTER_MIN_GROUPS; } else { - LASSERT(last_group >= FILTER_MIN_GROUPS); + LASSERT_MDS_GROUP(last_group); } CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name, @@ -1307,6 +1355,9 @@ static int filter_prep(struct obd_device *obd) GOTO(err_filp, rc = -EOPNOTSUPP); } + /** lu_target has very limited use in filter now */ + lut_init(NULL, &filter->fo_lut, obd, NULL); + rc = filter_init_server_data(obd, file); if (rc) { CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc); @@ -1369,7 +1420,7 @@ static void filter_post(struct obd_device *obd) if (rc) CERROR("error writing server data: rc = %d\n", rc); - for (i = 1; i < filter->fo_group_count; i++) { + for (i = 0; i < filter->fo_group_count; i++) { rc = filter_update_last_objid(obd, i, (i == filter->fo_group_count - 1)); if (rc) @@ -1416,7 +1467,6 @@ obd_id filter_last_id(struct filter_obd *filter, obd_gr group) spin_lock(&filter->fo_objidlock); id = filter->fo_last_objids[group]; spin_unlock(&filter->fo_objidlock); - return id; } @@ -1433,7 +1483,7 @@ struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid) struct filter_subdirs *subdirs; LASSERT(group < filter->fo_group_count); /* FIXME: object groups */ - if ((group > 0 && group < FILTER_GROUP_MDS0) || + if ((group > FILTER_GROUP_MDS0 && group < FILTER_GROUP_MDS1_N_BASE) || filter->fo_subdir_count == 0) return filter->fo_dentry_O_groups[group]; @@ -1531,9 +1581,8 @@ struct dentry *filter_fid2dentry(struct obd_device *obd, } static int filter_prepare_destroy(struct obd_device *obd, obd_id objid, - obd_id group) + obd_id group, struct lustre_handle *lockh) { - struct lustre_handle lockh; int flags = LDLM_AST_DISCARD_DATA, rc; struct ldlm_res_id res_id; ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } }; @@ -1545,15 +1594,19 @@ static int filter_prepare_destroy(struct obd_device *obd, obd_id objid, rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id, LDLM_EXTENT, &policy, LCK_PW, &flags, ldlm_blocking_ast, ldlm_completion_ast, NULL, NULL, 0, NULL, - &lockh); - - /* We only care about the side-effects, just drop the lock. */ - if (rc == ELDLM_OK) - ldlm_lock_decref(&lockh, LCK_PW); - + NULL, lockh); + if (rc != ELDLM_OK) + lockh->cookie = 0; RETURN(rc); } +static void filter_fini_destroy(struct obd_device *obd, + struct lustre_handle *lockh) +{ + if (lockh->cookie) + ldlm_lock_decref(lockh, LCK_PW); +} + /* This is vfs_unlink() without down(i_sem). If we call regular vfs_unlink() * we have 2.6 lock ordering issues with filter_commitrw_write() as it takes * i_sem before starting a handle, while filter_destroy() + vfs_unlink do the @@ -1889,6 +1942,30 @@ static void filter_iobuf_pool_done(struct filter_obd *filter) EXIT; } +static int filter_adapt_sptlrpc_conf(struct obd_device *obd, int initial) +{ + struct filter_obd *filter = &obd->u.filter; + struct sptlrpc_rule_set tmp_rset; + int rc; + + sptlrpc_rule_set_init(&tmp_rset); + rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial); + if (rc) { + CERROR("obd %s: failed get sptlrpc rules: %d\n", + obd->obd_name, rc); + return rc; + } + + sptlrpc_target_update_exp_flavor(obd, &tmp_rset); + + write_lock(&filter->fo_sptlrpc_lock); + sptlrpc_rule_set_free(&filter->fo_sptlrpc_rset); + filter->fo_sptlrpc_rset = tmp_rset; + write_unlock(&filter->fo_sptlrpc_lock); + + return 0; +} + /* * pre-allocate pool of iobuf's to be used by filter_{prep,commit}rw_write(). */ @@ -1915,7 +1992,8 @@ static int filter_iobuf_pool_init(struct filter_obd *filter) * If we haven't allocated a pool entry for this thread before, do so now. */ void *filter_iobuf_get(struct filter_obd *filter, struct obd_trans_info *oti) { - int thread_id = oti ? oti->oti_thread_id : -1; + int thread_id = (oti && oti->oti_thread) ? + oti->oti_thread->t_id : -1; struct filter_iobuf *pool = NULL; struct filter_iobuf **pool_place = NULL; @@ -1988,7 +2066,11 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, if (rc != 0) GOTO(err_ops, rc); - LASSERT(!lvfs_check_rdonly(lvfs_sbdev(mnt->mnt_sb))); + if (lvfs_check_rdonly(lvfs_sbdev(mnt->mnt_sb))) { + CERROR("%s: Underlying device is marked as read-only. " + "Setup failed\n", obd->obd_name); + GOTO(err_ops, rc = -EROFS); + } /* failover is the default */ obd->obd_replayable = 1; @@ -2031,7 +2113,7 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, sema_init(&filter->fo_alloc_lock, 1); init_brw_stats(&filter->fo_filter_stats); filter->fo_read_cache = 1; /* enable read-only cache by default */ - filter->fo_writethrough_cache = 0; /* disable writethrough cache */ + filter->fo_writethrough_cache = 1; /* enable writethrough cache */ filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE; filter->fo_fmd_max_num = FILTER_FMD_MAX_NUM_DEFAULT; filter->fo_fmd_max_age = FILTER_FMD_MAX_AGE_DEFAULT; @@ -2039,14 +2121,12 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, CFS_INIT_LIST_HEAD(&filter->fo_llog_list); spin_lock_init(&filter->fo_llog_list_lock); - rwlock_init(&filter->fo_sptlrpc_lock); - sptlrpc_rule_set_init(&filter->fo_sptlrpc_rset); + filter->fo_fl_oss_capa = 1; - filter->fo_fl_oss_capa = 0; CFS_INIT_LIST_HEAD(&filter->fo_capa_keys); filter->fo_capa_hash = init_capa_hash(); if (filter->fo_capa_hash == NULL) - GOTO(err_ops, rc = -ENOMEM); + GOTO(err_post, rc = -ENOMEM); sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid); obd->obd_namespace = ldlm_namespace_new(obd, ns_name, LDLM_NAMESPACE_SERVER, @@ -2066,6 +2146,11 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, GOTO(err_post, rc); } + rwlock_init(&filter->fo_sptlrpc_lock); + sptlrpc_rule_set_init(&filter->fo_sptlrpc_rset); + /* do this after llog being initialized */ + filter_adapt_sptlrpc_conf(obd, 1); + rc = lquota_setup(filter_quota_interface_ref, obd); if (rc) GOTO(err_post, rc); @@ -2092,17 +2177,14 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, if (obd->obd_recovering) { LCONSOLE_WARN("OST %s now serving %s (%s%s%s), but will be in " "recovery for at least %d:%.02d, or until %d " - "client%s reconnect. During this time new clients" - " will not be allowed to connect. " - "Recovery progress can be monitored by watching " - "/proc/fs/lustre/obdfilter/%s/recovery_status.\n", + "client%s reconnect%s.\n", obd->obd_name, lustre_cfg_string(lcfg, 1), label ?: "", label ? "/" : "", str, obd->obd_recovery_timeout / 60, obd->obd_recovery_timeout % 60, obd->obd_max_recoverable_clients, (obd->obd_max_recoverable_clients == 1) ? "":"s", - obd->obd_name); + (obd->obd_max_recoverable_clients == 1) ? "s":""); } else { LCONSOLE_INFO("OST %s now serving %s (%s%s%s) with recovery " "%s\n", obd->obd_name, lustre_cfg_string(lcfg, 1), @@ -2221,10 +2303,18 @@ static int filter_olg_fini(struct obd_llog_group *olg) rc = llog_cleanup(ctxt); ctxt = llog_group_get_ctxt(olg, LLOG_SIZE_ORIG_CTXT); - if (ctxt) + if (ctxt) { rc2 = llog_cleanup(ctxt); - if (!rc) - rc = rc2; + if (!rc) + rc = rc2; + } + + ctxt = llog_group_get_ctxt(olg, LLOG_CONFIG_ORIG_CTXT); + if (ctxt) { + rc2 = llog_cleanup(ctxt); + if (!rc) + rc = rc2; + } RETURN(rc); } @@ -2277,13 +2367,18 @@ filter_default_olg_init(struct obd_device *obd, struct obd_llog_group *olg, if (rc) GOTO(cleanup_lcm, rc); + rc = llog_setup(obd, olg, LLOG_CONFIG_ORIG_CTXT, tgt, 0, NULL, + &llog_lvfs_ops); + if (rc) + GOTO(cleanup_olg, rc); + ctxt = llog_group_get_ctxt(olg, LLOG_MDS_OST_REPL_CTXT); if (!ctxt) { CERROR("Can't get ctxt for %p:%x\n", olg, LLOG_MDS_OST_REPL_CTXT); GOTO(cleanup_olg, rc = -ENODEV); } - ctxt->loc_lcm = filter->fo_lcm; + ctxt->loc_lcm = lcm_get(filter->fo_lcm); ctxt->llog_proc_cb = filter_recov_log_mds_ost_cb; llog_ctxt_put(ctxt); @@ -2322,7 +2417,7 @@ filter_llog_init(struct obd_device *obd, struct obd_llog_group *olg, RETURN(-ENODEV); } ctxt->llog_proc_cb = filter_recov_log_mds_ost_cb; - ctxt->loc_lcm = filter->fo_lcm; + ctxt->loc_lcm = lcm_get(filter->fo_lcm); llog_ctxt_put(ctxt); RETURN(rc); } @@ -2334,23 +2429,33 @@ static int filter_llog_finish(struct obd_device *obd, int count) ENTRY; ctxt = llog_group_get_ctxt(&obd->obd_olg, LLOG_MDS_OST_REPL_CTXT); - LASSERT(ctxt != NULL); - mutex_down(&ctxt->loc_sem); - if (ctxt->loc_imp) { + if (ctxt) { /* - * Balance class_import_get() in llog_receptor_accept(). This - * is safe to do here, as llog is already synchronized and its - * import may go. + * Make sure that no cached llcds left in recov_thread. + * We actually do sync in disconnect time, but disconnect + * may not come being marked rq_no_resend = 1. */ - class_import_put(ctxt->loc_imp); - ctxt->loc_imp = NULL; + llog_sync(ctxt, NULL); + + /* + * Balance class_import_get() in llog_receptor_accept(). + * This is safe to do, as llog is already synchronized + * and its import may go. + */ + mutex_down(&ctxt->loc_sem); + if (ctxt->loc_imp) { + class_import_put(ctxt->loc_imp); + ctxt->loc_imp = NULL; + } + mutex_up(&ctxt->loc_sem); + llog_ctxt_put(ctxt); } - mutex_up(&ctxt->loc_sem); - llog_ctxt_put(ctxt); if (filter->fo_lcm) { + mutex_down(&ctxt->loc_sem); llog_recov_thread_fini(filter->fo_lcm, obd->obd_force); filter->fo_lcm = NULL; + mutex_up(&ctxt->loc_sem); } RETURN(filter_olg_fini(&obd->obd_olg)); } @@ -2474,6 +2579,9 @@ static int filter_llog_connect(struct obd_export *exp, obd->obd_name, body->lgdc_logid.lgl_oid, body->lgdc_logid.lgl_ogr, body->lgdc_logid.lgl_ogen); + spin_lock_bh(&obd->obd_processing_task_lock); + obd->u.filter.fo_mds_ost_sync = 1; + spin_unlock_bh(&obd->obd_processing_task_lock); rc = llog_connect(ctxt, &body->lgdc_logid, &body->lgdc_gen, NULL); llog_ctxt_put(ctxt); @@ -2530,7 +2638,8 @@ static int filter_precleanup(struct obd_device *obd, case OBD_CLEANUP_EARLY: break; case OBD_CLEANUP_EXPORTS: - target_cleanup_recovery(obd); + /* Stop recovery before namespace cleanup. */ + target_recovery_fini(obd); rc = filter_llog_preclean(obd); break; } @@ -2546,14 +2655,8 @@ static int filter_cleanup(struct obd_device *obd) LCONSOLE_WARN("%s: shutting down for failover; client state " "will be preserved.\n", obd->obd_name); - if (!list_empty(&obd->obd_exports)) { - CERROR("%s: still has clients!\n", obd->obd_name); - class_disconnect_exports(obd); - if (!list_empty(&obd->obd_exports)) { - CERROR("still has exports after forced cleanup?\n"); - RETURN(-EBUSY); - } - } + obd_exports_barrier(obd); + obd_zombie_barrier(); lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry); lprocfs_free_per_client_stats(obd); @@ -2561,10 +2664,6 @@ static int filter_cleanup(struct obd_device *obd) lprocfs_obd_cleanup(obd); lquota_cleanup(filter_quota_interface_ref, obd); - /* Stop recovery before namespace cleanup. */ - target_stop_recovery_thread(obd); - target_cleanup_recovery(obd); - ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force); obd->obd_namespace = NULL; @@ -2618,7 +2717,13 @@ static int filter_connect_internal(struct obd_export *exp, exp->exp_connect_flags = data->ocd_connect_flags; data->ocd_version = LUSTRE_VERSION_CODE; + /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */ + if (!ergo(data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN, + data->ocd_connect_flags & OBD_CONNECT_MDS)) + RETURN(-EPROTO); + if (exp->exp_connect_flags & OBD_CONNECT_GRANT) { + struct filter_obd *filter = &exp->exp_obd->u.filter; obd_size left, want; spin_lock(&exp->exp_obd->obd_osfs_lock); @@ -2632,6 +2737,8 @@ static int filter_connect_internal(struct obd_export *exp, LPU64" left: "LPU64"\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp, data->ocd_grant, want, left); + + filter->fo_tot_granted_clients ++; } if (data->ocd_connect_flags & OBD_CONNECT_INDEX) { @@ -2655,6 +2762,10 @@ static int filter_connect_internal(struct obd_export *exp, data->ocd_index); RETURN(-EBADF); } + /* FIXME: Do the same with the MDS UUID and fsd_peeruuid. + * FIXME: We don't strictly need the COMPAT flag for that, + * FIXME: as fsd_peeruuid[0] will tell us if that is set. + * FIXME: We needed it for the index, as index 0 is valid. */ } if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) { @@ -2689,11 +2800,6 @@ static int filter_connect_internal(struct obd_export *exp, obd_export_nid2str(exp)); } - /* FIXME: Do the same with the MDS UUID and fsd_peeruuid. - * FIXME: We don't strictly need the COMPAT flag for that, - * FIXME: as fsd_peeruuid[0] will tell us if that is set. - * FIXME: We needed it for the index, as index 0 is valid. */ - RETURN(0); } @@ -2718,34 +2824,35 @@ static int filter_reconnect(const struct lu_env *env, /* nearly identical to mds_connect */ static int filter_connect(const struct lu_env *env, - struct lustre_handle *conn, struct obd_device *obd, + struct obd_export **exp, struct obd_device *obd, struct obd_uuid *cluuid, struct obd_connect_data *data, void *localdata) { struct lvfs_run_ctxt saved; - struct obd_export *exp; + struct lustre_handle conn = { 0 }; + struct obd_export *lexp; struct filter_export_data *fed; struct lsd_client_data *lcd = NULL; __u32 group; int rc; ENTRY; - if (conn == NULL || obd == NULL || cluuid == NULL) + if (exp == NULL || obd == NULL || cluuid == NULL) RETURN(-EINVAL); - rc = class_connect(conn, obd, cluuid); + rc = class_connect(&conn, obd, cluuid); if (rc) RETURN(rc); - exp = class_conn2export(conn); - LASSERT(exp != NULL); + lexp = class_conn2export(&conn); + LASSERT(lexp != NULL); - fed = &exp->exp_filter_data; + fed = &lexp->exp_filter_data; - rc = filter_connect_internal(exp, data); + rc = filter_connect_internal(lexp, data); if (rc) GOTO(cleanup, rc); - filter_export_stats_init(obd, exp, localdata); + filter_export_stats_init(obd, lexp, localdata); if (obd->obd_replayable) { OBD_ALLOC(lcd, sizeof(*lcd)); if (!lcd) { @@ -2755,17 +2862,15 @@ static int filter_connect(const struct lu_env *env, memcpy(lcd->lcd_uuid, cluuid, sizeof(lcd->lcd_uuid)); fed->fed_lcd = lcd; - rc = filter_client_add(obd, exp, -1); + rc = filter_client_add(obd, lexp, -1); if (rc) GOTO(cleanup, rc); } group = data->ocd_group; - if (group == 0) - GOTO(cleanup, rc); CWARN("%s: Received MDS connection ("LPX64"); group %d\n", - obd->obd_name, exp->exp_handle.h_cookie, group); + obd->obd_name, lexp->exp_handle.h_cookie, group); push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); rc = filter_read_groups(obd, group, 1); @@ -2783,9 +2888,10 @@ cleanup: OBD_FREE_PTR(lcd); fed->fed_lcd = NULL; } - class_disconnect(exp); + class_disconnect(lexp); + *exp = NULL; } else { - class_export_put(exp); + *exp = lexp; } RETURN(rc); @@ -2912,9 +3018,7 @@ static int filter_destroy_export(struct obd_export *exp) exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp, exp->exp_filter_data.fed_pending); - /* Not ported yet the b1_6 quota functionality - * lquota_clearinfo(filter_quota_interface_ref, exp, exp->exp_obd); - */ + lquota_clearinfo(filter_quota_interface_ref, exp, exp->exp_obd); target_destroy_export(exp); ldlm_destroy_export(exp); @@ -2931,6 +3035,12 @@ static int filter_destroy_export(struct obd_export *exp) filter_grant_discard(exp); filter_fmd_cleanup(exp); + if (exp->exp_connect_flags & OBD_CONNECT_GRANT_SHRINK) { + struct filter_obd *filter = &exp->exp_obd->u.filter; + if (filter->fo_tot_granted_clients > 0) + filter->fo_tot_granted_clients --; + } + if (!(exp->exp_flags & OBD_OPT_FORCE)) filter_grant_sanity_check(exp->exp_obd, __FUNCTION__); @@ -2941,7 +3051,7 @@ static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp) { struct obd_llog_group *olg_min, *olg; struct filter_obd *filter; - int worked = 0, group; + int worked = -1, group; struct llog_ctxt *ctxt; ENTRY; @@ -2978,18 +3088,20 @@ static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp) (dexp == olg_min->olg_exp || dexp == NULL)) { int err; ctxt = llog_group_get_ctxt(olg_min, - LLOG_MDS_OST_REPL_CTXT); - LASSERT(ctxt != NULL); - err = llog_sync(ctxt, olg_min->olg_exp); - llog_ctxt_put(ctxt); - if (err) - CERROR("error flushing logs to MDS: rc %d\n", - err); + LLOG_MDS_OST_REPL_CTXT); + if (ctxt) { + err = llog_sync(ctxt, olg_min->olg_exp); + llog_ctxt_put(ctxt); + if (err) { + CERROR("error flushing logs to MDS: " + "rc %d\n", err); + } + } } } while (olg_min != NULL); } -/* also incredibly similar to mds_disconnect */ +/* Also incredibly similar to mds_disconnect */ static int filter_disconnect(struct obd_export *exp) { struct obd_device *obd = exp->exp_obd; @@ -3003,6 +3115,11 @@ static int filter_disconnect(struct obd_export *exp) filter_grant_sanity_check(obd, __FUNCTION__); filter_grant_discard(exp); + /* Flush any remaining cancel messages out to the target */ + filter_sync_llogs(obd, exp); + + lquota_clearinfo(filter_quota_interface_ref, exp, exp->exp_obd); + /* Disconnect early so that clients can't keep using export */ rc = class_disconnect(exp); if (exp->exp_obd->obd_namespace != NULL) @@ -3012,8 +3129,6 @@ static int filter_disconnect(struct obd_export *exp) lprocfs_exp_cleanup(exp); - /* flush any remaining cancel messages out to the target */ - filter_sync_llogs(obd, exp); class_export_put(exp); RETURN(rc); } @@ -3035,7 +3150,6 @@ static void filter_revimp_update(struct obd_export *exp) static int filter_ping(struct obd_export *exp) { filter_fmd_expire(exp); - return 0; } @@ -3141,7 +3255,8 @@ int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry, unsigned int orig_ids[MAXQUOTAS] = {0, 0}; struct llog_cookie *fcc = NULL; struct filter_obd *filter; - int rc, err, locked = 0, sync = 0; + int rc, err, sync = 0; + loff_t old_size = 0; unsigned int ia_valid; struct inode *inode; struct iattr iattr; @@ -3163,13 +3278,22 @@ int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry, if (fcc != NULL) *fcc = oa->o_lcookie; } - - if (ia_valid & ATTR_SIZE || ia_valid & (ATTR_UID | ATTR_GID)) { + if (ia_valid & (ATTR_SIZE | ATTR_UID | ATTR_GID)) { DQUOT_INIT(inode); + /* Filter truncates and writes are serialized by + * i_alloc_sem, see the comment in + * filter_preprw_write.*/ + if (ia_valid & ATTR_SIZE) + down_write(&inode->i_alloc_sem); LOCK_INODE_MUTEX(inode); - locked = 1; + old_size = i_size_read(inode); } + /* VBR: version recovery check */ + rc = filter_version_get_check(exp, oti, inode); + if (rc) + GOTO(out_unlock, rc); + /* If the inode still has SUID+SGID bits set (see filter_precreate()) * then we will accept the UID+GID sent by the client during write for * initializing the ownership of this inode. We only allow this to @@ -3232,7 +3356,7 @@ int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry, * sure we have one left for the last_rcvd update. */ err = fsfilt_extend(exp->exp_obd, inode, 1, handle); - rc = filter_finish_transno(exp, oti, rc, sync); + rc = filter_finish_transno(exp, inode, oti, rc, sync); if (sync) { filter_cancel_cookies_cb(exp->exp_obd, 0, fcc, rc); fcc = NULL; @@ -3247,18 +3371,22 @@ int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry, fcc = NULL; } - if (locked) { - /* truncate can leave dirty pages in the cache. - * we'll take care of them in write path -bzzz */ - UNLOCK_INODE_MUTEX(inode); - locked = 0; + /* For a partial-page truncate flush the page to disk immediately + * to avoid data corruption during direct disk write. b=17397 */ + if (!sync && (iattr.ia_valid & ATTR_SIZE) && + old_size != iattr.ia_size && (iattr.ia_size & ~CFS_PAGE_MASK)) { + err = filemap_fdatawrite_range(inode->i_mapping, iattr.ia_size, + iattr.ia_size + 1); + if (!rc) + rc = err; } EXIT; out_unlock: - if (locked) + if (ia_valid & (ATTR_SIZE | ATTR_UID | ATTR_GID)) UNLOCK_INODE_MUTEX(inode); - + if (ia_valid & ATTR_SIZE) + up_write(&inode->i_alloc_sem); if (fcc) OBD_FREE(fcc, sizeof(*fcc)); @@ -3278,43 +3406,52 @@ out_unlock: int filter_setattr(struct obd_export *exp, struct obd_info *oinfo, struct obd_trans_info *oti) { + struct obdo *oa = oinfo->oi_oa; + struct lustre_capa *capa = oinfo_capa(oinfo); struct ldlm_res_id res_id; struct filter_mod_data *fmd; struct lvfs_run_ctxt saved; struct filter_obd *filter; struct ldlm_resource *res; struct dentry *dentry; + __u64 opc = CAPA_OPC_META_WRITE; int rc; ENTRY; - osc_build_res_name(oinfo->oi_oa->o_id, oinfo->oi_oa->o_gr, &res_id); - rc = filter_auth_capa(exp, NULL, oinfo_mdsno(oinfo), - oinfo_capa(oinfo), CAPA_OPC_META_WRITE); + if (oa->o_valid & OBD_FL_TRUNC) + opc |= CAPA_OPC_OSS_TRUNC; + rc = filter_auth_capa(exp, NULL, obdo_mdsno(oa), capa, opc); if (rc) RETURN(rc); + if (oa->o_valid & (OBD_MD_FLUID | OBD_MD_FLGID)) { + rc = filter_capa_fixoa(exp, oa, obdo_mdsno(oa), capa); + if (rc) + RETURN(rc); + } + + osc_build_res_name(oa->o_id, oa->o_gr, &res_id); /* This would be very bad - accidentally truncating a file when * changing the time or similar - bug 12203. */ - if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE && + if (oa->o_valid & OBD_MD_FLSIZE && oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) { static char mdsinum[48]; - if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) + if (oa->o_valid & OBD_MD_FLFID) snprintf(mdsinum, sizeof(mdsinum) - 1, - " of inode "LPU64"/%u", oinfo->oi_oa->o_fid, - oinfo->oi_oa->o_generation); + " of inode "LPU64"/%u", oa->o_fid, + oa->o_generation); else mdsinum[0] = '\0'; CERROR("%s: setattr from %s trying to truncate objid "LPU64 " %s\n", exp->exp_obd->obd_name, obd_export_nid2str(exp), - oinfo->oi_oa->o_id, mdsinum); + oa->o_id, mdsinum); RETURN(-EPERM); } - dentry = __filter_oa2dentry(exp->exp_obd, oinfo->oi_oa, - __FUNCTION__, 1); + dentry = __filter_oa2dentry(exp->exp_obd, oa, __FUNCTION__, 1); if (IS_ERR(dentry)) RETURN(PTR_ERR(dentry)); @@ -3322,16 +3459,16 @@ int filter_setattr(struct obd_export *exp, struct obd_info *oinfo, push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL); lock_kernel(); - if (oinfo->oi_oa->o_valid & + if (oa->o_valid & (OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME)) { - fmd = filter_fmd_get(exp,oinfo->oi_oa->o_id,oinfo->oi_oa->o_gr); + fmd = filter_fmd_get(exp, oa->o_id, oa->o_gr); if (fmd && fmd->fmd_mactime_xid < oti->oti_xid) fmd->fmd_mactime_xid = oti->oti_xid; filter_fmd_put(exp, fmd); } /* setting objects attributes (including owner/group) */ - rc = filter_setattr_internal(exp, dentry, oinfo->oi_oa, oti); + rc = filter_setattr_internal(exp, dentry, oa, oti); if (rc) GOTO(out_unlock, rc); @@ -3345,10 +3482,10 @@ int filter_setattr(struct obd_export *exp, struct obd_info *oinfo, ldlm_resource_putref(res); } - oinfo->oi_oa->o_valid = OBD_MD_FLID; + oa->o_valid = OBD_MD_FLID; /* Quota release need uid/gid info */ - obdo_from_inode(oinfo->oi_oa, dentry->d_inode, + obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS | OBD_MD_FLUID | OBD_MD_FLGID); EXIT; @@ -3417,15 +3554,16 @@ static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp, /* caller must hold fo_create_locks[oa->o_gr] */ static int filter_destroy_precreated(struct obd_export *exp, struct obdo *oa, - struct filter_obd *filter) + struct filter_obd *filter) { - struct obdo doa; /* XXX obdo on stack */ + struct obdo doa = { 0 }; /* XXX obdo on stack */ obd_id last, id; - int rc; + int rc = 0; + int skip_orphan; ENTRY; LASSERT(oa); - LASSERT(oa->o_gr != 0); + LASSERT_MDS_GROUP(oa->o_gr); LASSERT(oa->o_valid & OBD_MD_FLGROUP); LASSERT(down_trylock(&filter->fo_create_locks[oa->o_gr]) != 0); @@ -3443,26 +3581,38 @@ static int filter_destroy_precreated(struct obd_export *exp, struct obdo *oa, last = filter_last_id(filter, doa.o_gr); - CWARN("%s: deleting orphan objects from "LPU64" to "LPU64"\n", - exp->exp_obd->obd_name, oa->o_id + 1, last); + skip_orphan = !!(exp->exp_connect_flags & OBD_CONNECT_SKIP_ORPHAN); + + CDEBUG(D_HA, "%s: deleting orphan objects from "LPU64" to "LPU64"%s\n", + exp->exp_obd->obd_name, oa->o_id + 1, last, + skip_orphan ? ", orphan objids won't be reused any more." : "."); for (id = last; id > oa->o_id; id--) { doa.o_id = id; - rc = filter_destroy(exp, &doa, NULL, NULL, NULL); + rc = filter_destroy(exp, &doa, NULL, NULL, NULL, NULL); if (rc && rc != -ENOENT) /* this is pretty fatal... */ CEMERG("error destroying precreate objid "LPU64": %d\n", id, rc); - filter_set_last_id(filter, id - 1, doa.o_gr); + /* update last_id on disk periodically so that if we restart * we don't need to re-scan all of the just-deleted objects. */ - if ((id & 511) == 0) + if ((id & 511) == 0 && !skip_orphan) { + filter_set_last_id(filter, id - 1, doa.o_gr); filter_update_last_objid(exp->exp_obd, doa.o_gr, 0); + } } CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n", exp->exp_obd->obd_name, doa.o_gr, oa->o_id); - rc = filter_update_last_objid(exp->exp_obd, doa.o_gr, 1); + if (!skip_orphan) { + filter_set_last_id(filter, id, doa.o_gr); + rc = filter_update_last_objid(exp->exp_obd, doa.o_gr, 1); + } else { + /* don't reuse orphan object, return last used objid */ + oa->o_id = last; + rc = 0; + } clear_bit(doa.o_gr, &filter->fo_destroys_in_progress); RETURN(rc); @@ -3481,6 +3631,8 @@ static int filter_handle_precreate(struct obd_export *exp, struct obdo *oa, /* delete orphans request */ if ((oa->o_valid & OBD_MD_FLFLAGS) && (oa->o_flags & OBD_FL_DELORPHAN)){ + obd_id last = filter_last_id(filter, group); + if (oti->oti_conn_cnt < exp->exp_conn_cnt) { CERROR("%s: dropping old orphan cleanup request\n", obd->obd_name); @@ -3495,14 +3647,14 @@ static int filter_handle_precreate(struct obd_export *exp, struct obdo *oa, up(&filter->fo_create_locks[group]); RETURN(0); } - diff = oa->o_id - filter_last_id(filter, group); + diff = oa->o_id - last; CDEBUG(D_HA, "filter_last_id() = "LPU64" -> diff = %d\n", - filter_last_id(filter, group), diff); + last, diff); if (-diff > OST_MAX_PRECREATE) { CERROR("%s: ignoring bogus orphan destroy request: " "obdid "LPU64" last_id "LPU64"\n", obd->obd_name, - oa->o_id, filter_last_id(filter, group)); + oa->o_id, last); /* FIXME: should reset precreate_next_id on MDS */ GOTO(out, rc = -EINVAL); } @@ -3523,8 +3675,8 @@ static int filter_handle_precreate(struct obd_export *exp, struct obdo *oa, obd->obd_name); GOTO(out, rc = 0); } - /* only precreate if group == 0 and o_id is specified */ - if (group < FILTER_GROUP_MDS0 || oa->o_id == 0) + /* only precreate if group == 0 and o_id is specfied */ + if (group == FILTER_GROUP_LLOG || oa->o_id == 0) diff = 1; else diff = oa->o_id - filter_last_id(filter, group); @@ -3582,7 +3734,7 @@ static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs, * stop creating files on MDS if OST is not good shape to create * objects.*/ osfs->os_state = (filter->fo_obt.obt_sb->s_flags & MS_RDONLY) ? - EROFS : 0; + OS_STATE_READONLY : 0; RETURN(rc); } @@ -3610,6 +3762,19 @@ static int filter_use_existing_obj(struct obd_device *obd, return rc; } +static __u64 filter_calc_free_inodes(struct obd_device *obd) +{ + int rc; + __u64 os_ffree = -1; + + spin_lock(&obd->obd_osfs_lock); + rc = fsfilt_statfs(obd, obd->u.obt.obt_sb, cfs_time_shift_64(1)); + if (rc == 0) + os_ffree = obd->obd_osfs.os_ffree; + spin_unlock(&obd->obd_osfs_lock); + + return os_ffree; +} /* We rely on the fact that only one thread will be creating files in a given * group at a time, which is why we don't need an atomic filter_get_new_id. @@ -3630,6 +3795,7 @@ static int filter_precreate(struct obd_device *obd, struct obdo *oa, struct obd_statfs *osfs; int err = 0, rc = 0, recreate_obj = 0, i; cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT/2); + __u64 os_ffree; obd_id next_id; void *handle = NULL; ENTRY; @@ -3742,9 +3908,19 @@ static int filter_precreate(struct obd_device *obd, struct obdo *oa, S_IFREG | S_ISUID | S_ISGID | 0666, NULL); if (rc) { CERROR("create failed rc = %d\n", rc); + if (rc == -ENOSPC) { + os_ffree = filter_calc_free_inodes(obd); + if (os_ffree != -1) + CERROR("%s: free inode "LPU64"\n", + obd->obd_name, os_ffree); + } GOTO(cleanup, rc); } + if (dchild->d_inode) + CDEBUG(D_INFO, "objid "LPU64" got inum %lu\n", next_id, + dchild->d_inode->i_ino); + set_last_id: if (!recreate_obj) { filter_set_last_id(filter, next_id, group); @@ -3803,7 +3979,7 @@ static int filter_create(struct obd_export *exp, struct obdo *oa, CDEBUG(D_INODE, "%s: filter_create(od->o_gr="LPU64",od->o_id=" LPU64")\n", obd->obd_name, oa->o_gr, oa->o_id); - if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) { + if (!(oa->o_valid & OBD_MD_FLGROUP)) { CERROR("!!! nid %s sent invalid object group %d\n", obd_export_nid2str(exp), group); RETURN(-EINVAL); @@ -3867,12 +4043,13 @@ static int filter_create(struct obd_export *exp, struct obdo *oa, int filter_destroy(struct obd_export *exp, struct obdo *oa, struct lov_stripe_md *md, struct obd_trans_info *oti, - struct obd_export *md_exp) + struct obd_export *md_exp, void *capa) { unsigned int qcids[MAXQUOTAS] = {0, 0}; struct obd_device *obd; struct filter_obd *filter; struct dentry *dchild = NULL, *dparent = NULL; + struct lustre_handle lockh = { 0 }; struct lvfs_run_ctxt saved; void *handle = NULL; struct llog_cookie *fcc = NULL; @@ -3882,6 +4059,11 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, LASSERT(oa->o_valid & OBD_MD_FLGROUP); + rc = filter_auth_capa(exp, NULL, obdo_mdsno(oa), + (struct lustre_capa *)capa, CAPA_OPC_OSS_DESTROY); + if (rc) + RETURN(rc); + obd = exp->exp_obd; filter = &obd->u.filter; @@ -3920,7 +4102,7 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, GOTO(cleanup, rc = -ENOENT); } - filter_prepare_destroy(obd, oa->o_id, oa->o_gr); + filter_prepare_destroy(obd, oa->o_id, oa->o_gr, &lockh); /* Our MDC connection is established by the MDS to us */ if (oa->o_valid & OBD_MD_FLCOOKIE) { @@ -3938,6 +4120,12 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, * (see BUG 4180) -bzzz */ LOCK_INODE_MUTEX(dchild->d_inode); + + /* VBR: version recovery check */ + rc = filter_version_get_check(exp, oti, dchild->d_inode); + if (rc) + GOTO(cleanup, rc); + handle = fsfilt_start_log(obd, dchild->d_inode, FSFILT_OP_SETATTR, NULL, 1); if (IS_ERR(handle)) { @@ -3994,7 +4182,7 @@ cleanup: * on commit. then we call callback directly to free * the fcc. */ - rc = filter_finish_transno(exp, oti, rc, sync); + rc = filter_finish_transno(exp, NULL, oti, rc, sync); if (sync) { filter_cancel_cookies_cb(obd, 0, fcc, rc); fcc = NULL; @@ -4010,6 +4198,8 @@ cleanup: case 3: filter_parent_unlock(dparent); case 2: + filter_fini_destroy(obd, &lockh); + f_dput(dchild); if (fcc != NULL) OBD_FREE(fcc, sizeof(*fcc)); @@ -4026,9 +4216,8 @@ cleanup: qcids[GRPQUOTA] = oa->o_gid; rc2 = lquota_adjust(filter_quota_interface_ref, obd, qcids, NULL, rc, FSFILT_OP_UNLINK); - if (rc2) - CDEBUG(D_QUOTA, "filter adjust qunit! (rc:%d)\n", rc2); + CERROR("filter adjust qunit! (rc:%d)\n", rc2); return rc; } @@ -4050,13 +4239,10 @@ static int filter_truncate(struct obd_export *exp, struct obd_info *oinfo, ", o_size = "LPD64"\n", oinfo->oi_oa->o_id, oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start); - rc = filter_auth_capa(exp, NULL, oinfo_mdsno(oinfo), - oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC); - if (rc) - RETURN(rc); - oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start; + oinfo->oi_oa->o_valid |= OBD_FL_TRUNC; rc = filter_setattr(exp, oinfo, oti); + oinfo->oi_oa->o_valid &= ~OBD_FL_TRUNC; RETURN(rc); } @@ -4077,10 +4263,10 @@ static int filter_sync(struct obd_export *exp, struct obdo *oa, filter = &exp->exp_obd->u.filter; - /* an objid of zero is taken to mean "sync whole filesystem" */ + /* An objid of zero is taken to mean "sync whole filesystem" */ if (!oa || !(oa->o_valid & OBD_MD_FLID)) { rc = fsfilt_sync(exp->exp_obd, filter->fo_obt.obt_sb); - /* flush any remaining cancel messages out to the target */ + /* Flush any remaining cancel messages out to the target */ filter_sync_llogs(exp->exp_obd, exp); RETURN(rc); } @@ -4186,28 +4372,92 @@ static int filter_get_info(struct obd_export *exp, __u32 keylen, push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); rc = fsfilt_iocontrol(obd, dentry->d_inode, NULL, EXT3_IOC_FIEMAP, (long)fiemap); - if (rc) { - f_dput(dentry); - RETURN(rc); - } pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); f_dput(dentry); - RETURN(0); + RETURN(rc); } CDEBUG(D_IOCTL, "invalid key\n"); RETURN(-EINVAL); } +static inline int filter_setup_llog_group(struct obd_export *exp, + struct obd_device *obd, + int group) +{ + struct obd_llog_group *olg; + struct llog_ctxt *ctxt; + int rc; + + olg = filter_find_create_olg(obd, group); + if (IS_ERR(olg)) + RETURN(PTR_ERR(olg)); + + llog_group_set_export(olg, exp); + + ctxt = llog_group_get_ctxt(olg, LLOG_MDS_OST_REPL_CTXT); + LASSERTF(ctxt != NULL, "ctxt is null\n"); + + rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse); + llog_ctxt_put(ctxt); + return rc; +} + +static int filter_set_grant_shrink(struct obd_export *exp, + struct ost_body *body) +{ + /* handle shrink grant */ + spin_lock(&exp->exp_obd->obd_osfs_lock); + filter_grant_incoming(exp, &body->oa); + spin_unlock(&exp->exp_obd->obd_osfs_lock); + + RETURN(0); + +} + +static int filter_set_mds_conn(struct obd_export *exp, void *val) +{ + struct obd_device *obd; + int rc = 0, group; + ENTRY; + + obd = exp->exp_obd; + if (obd == NULL) { + CDEBUG(D_IOCTL, "invalid export %p\n", exp); + RETURN(-EINVAL); + } + + LCONSOLE_WARN("%s: received MDS connection from %s\n", obd->obd_name, + obd_export_nid2str(exp)); + obd->u.filter.fo_mdc_conn.cookie = exp->exp_handle.h_cookie; + + /* setup llog imports */ + if (val != NULL) + group = (int)(*(__u32 *)val); + else + group = 0; /* default value */ + + LASSERT_MDS_GROUP(group); + rc = filter_setup_llog_group(exp, obd, group); + if (rc) + goto out; + + if (group == FILTER_GROUP_MDS0) { + /* setup llog group 1 for interop */ + filter_setup_llog_group(exp, obd, FILTER_GROUP_LLOG); + } + + lquota_setinfo(filter_quota_interface_ref, obd, exp); +out: + RETURN(rc); +} + static int filter_set_info_async(struct obd_export *exp, __u32 keylen, void *key, __u32 vallen, void *val, struct ptlrpc_request_set *set) { struct obd_device *obd; - struct obd_llog_group *olg; - struct llog_ctxt *ctxt; - int rc = 0, group; ENTRY; obd = exp->exp_obd; @@ -4217,6 +4467,7 @@ static int filter_set_info_async(struct obd_export *exp, __u32 keylen, } if (KEY_IS(KEY_CAPA_KEY)) { + int rc; rc = filter_update_capa_key(obd, (struct lustre_capa_key *)val); if (rc) CERROR("filter update capability key failed: %d\n", rc); @@ -4225,36 +4476,22 @@ static int filter_set_info_async(struct obd_export *exp, __u32 keylen, if (KEY_IS(KEY_REVIMP_UPD)) { filter_revimp_update(exp); + lquota_clearinfo(filter_quota_interface_ref, exp, exp->exp_obd); RETURN(0); } - if (!KEY_IS(KEY_MDS_CONN)) - RETURN(-EINVAL); - - LCONSOLE_WARN("%s: received MDS connection from %s\n", obd->obd_name, - obd_export_nid2str(exp)); - obd->u.filter.fo_mdc_conn.cookie = exp->exp_handle.h_cookie; - - /* setup llog imports */ - LASSERT(val != NULL); - group = (int)(*(__u32 *)val); - LASSERT(group >= FILTER_GROUP_MDS0); - - olg = filter_find_create_olg(obd, group); - if (IS_ERR(olg)) - RETURN(PTR_ERR(olg)); - - llog_group_set_export(olg, exp); - - ctxt = llog_group_get_ctxt(olg, LLOG_MDS_OST_REPL_CTXT); - LASSERTF(ctxt != NULL, "ctxt is null\n"); + if (KEY_IS(KEY_SPTLRPC_CONF)) { + filter_adapt_sptlrpc_conf(obd, 0); + RETURN(0); + } - rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse); - llog_ctxt_put(ctxt); + if (KEY_IS(KEY_MDS_CONN)) + RETURN(filter_set_mds_conn(exp, val)); - lquota_setinfo(filter_quota_interface_ref, exp, obd); + if (KEY_IS(KEY_GRANT_SHRINK)) + RETURN(filter_set_grant_shrink(exp, val)); - RETURN(rc); + RETURN(-EINVAL); } int filter_iocontrol(unsigned int cmd, struct obd_export *exp, @@ -4266,7 +4503,7 @@ int filter_iocontrol(unsigned int cmd, struct obd_export *exp, switch (cmd) { case OBD_IOC_ABORT_RECOVERY: { - CERROR("aborting recovery for device %s\n", obd->obd_name); + LCONSOLE_WARN("%s: Aborting recovery.\n", obd->obd_name); target_stop_recovery_thread(obd); RETURN(0); } @@ -4360,39 +4597,13 @@ static int filter_process_config(struct obd_device *obd, obd_count len, int rc = 0; switch (lcfg->lcfg_command) { - case LCFG_SPTLRPC_CONF: { - struct filter_obd *filter = &obd->u.filter; - struct sptlrpc_conf_log *log; - struct sptlrpc_rule_set tmp_rset; - - log = sptlrpc_conf_log_extract(lcfg); - if (IS_ERR(log)) { - rc = PTR_ERR(log); - break; - } - - sptlrpc_rule_set_init(&tmp_rset); - - rc = sptlrpc_rule_set_from_log(&tmp_rset, log); - if (rc) { - CERROR("obd %s: failed get sptlrpc rules: %d\n", - obd->obd_name, rc); - break; - } - - write_lock(&filter->fo_sptlrpc_lock); - sptlrpc_rule_set_free(&filter->fo_sptlrpc_rset); - filter->fo_sptlrpc_rset = tmp_rset; - write_unlock(&filter->fo_sptlrpc_lock); - - sptlrpc_target_update_exp_flavor(obd, &tmp_rset); - break; - } default: lprocfs_filter_init_vars(&lvars); rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, lcfg, obd); + if (rc > 0) + rc = 0; break; }