X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdfilter%2Ffilter.c;h=13c2530618bc3eee320b65974713eb06fe40e49a;hp=9c346bfe3efc82859cd1bbe9b9a9f380a016f9d1;hb=e67c6e366752611ffd2baeb7cefa24c9f289eb78;hpb=2d594879dd55133ee0403c3134b4ef2397615897 diff --git a/lustre/obdfilter/filter.c b/lustre/obdfilter/filter.c index 9c346bf..13c2530 100644 --- a/lustre/obdfilter/filter.c +++ b/lustre/obdfilter/filter.c @@ -79,24 +79,50 @@ #include "filter_internal.h" -/* Group 0 is no longer a legal group, to catch uninitialized IDs */ -#define FILTER_MIN_GROUPS FILTER_GROUP_MDS0 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); + class_export_cb_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); + cfs_spin_lock(&exp->exp_lock); + exp->exp_vbr_failed = 1; + cfs_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; - struct lsd_client_data *lcd = fed->fed_lcd; + struct lsd_client_data *lcd; __u64 last_rcvd; loff_t off; int err, log_pri = D_RPCTRACE; @@ -108,25 +134,41 @@ int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti, if (!exp->exp_obd->obd_replayable || oti == NULL) RETURN(rc); + cfs_mutex_down(&fed->fed_lcd_lock); + lcd = fed->fed_lcd; + /* if the export has already been disconnected, we have no last_rcvd slot, + * update server data with latest transno then */ + if (lcd == NULL) { + cfs_mutex_up(&fed->fed_lcd_lock); + CWARN("commit transaction for disconnected client %s: rc %d\n", + exp->exp_client_uuid.uuid, rc); + err = filter_update_server_data(exp->exp_obd, + filter->fo_rcvd_filp, + filter->fo_fsd); + RETURN(err); + } + /* we don't allocate new transnos for replayed requests */ + cfs_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); - spin_unlock(&filter->fo_translock); } + oti->oti_transno = last_rcvd; + + LASSERT(last_rcvd >= le64_to_cpu(lcd->lcd_last_transno)); 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); + cfs_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 +176,19 @@ 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_cb_get(exp); /* released when the cb is called */ 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; @@ -155,7 +198,7 @@ int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti, CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n", last_rcvd, lcd->lcd_uuid, fed->fed_lr_idx, err); - + cfs_mutex_up(&fed->fed_lcd_lock); RETURN(rc); } @@ -173,7 +216,7 @@ static void init_brw_stats(struct brw_stats *brw_stats) { int i; for (i = 0; i < BRW_LAST; i++) - spin_lock_init(&brw_stats->hist[i].oh_lock); + cfs_spin_lock_init(&brw_stats->hist[i].oh_lock); } static int lprocfs_init_rw_stats(struct obd_device *obd, @@ -183,7 +226,7 @@ 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; @@ -202,19 +245,21 @@ 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); rc = lprocfs_exp_setup(exp, client_nid, &newnid); - if (rc) + if (rc) { + /* Mask error for already created + * /proc entries */ + if (rc == -EALREADY) + rc = 0; RETURN(rc); + } if (newnid) { struct nid_stat *tmp = exp->exp_nid_stats; @@ -222,7 +267,7 @@ static int filter_export_stats_init(struct obd_device *obd, OBD_ALLOC(tmp->nid_brw_stats, sizeof(struct brw_stats)); if (tmp->nid_brw_stats == NULL) - RETURN(-ENOMEM); + GOTO(clean, rc = -ENOMEM); init_brw_stats(tmp->nid_brw_stats); rc = lprocfs_seq_create(exp->exp_nid_stats->nid_proc, "brw_stats", @@ -233,15 +278,30 @@ static int filter_export_stats_init(struct obd_device *obd, rc = lprocfs_init_rw_stats(obd, &exp->exp_nid_stats->nid_stats); if (rc) - RETURN(rc); + GOTO(clean, rc); rc = lprocfs_register_stats(tmp->nid_proc, "stats", tmp->nid_stats); if (rc) - RETURN(rc); + GOTO(clean, rc); + /* Always add in ldlm_stats */ + tmp->nid_ldlm_stats = + lprocfs_alloc_stats(LDLM_LAST_OPC - LDLM_FIRST_OPC, + LPROCFS_STATS_FLAG_NOPERCPU); + if (tmp->nid_ldlm_stats == NULL) + GOTO(clean, rc = -ENOMEM); + + lprocfs_init_ldlm_stats(tmp->nid_ldlm_stats); + rc = lprocfs_register_stats(tmp->nid_proc, "ldlm_stats", + tmp->nid_ldlm_stats); + if (rc) + GOTO(clean, rc); } RETURN(0); + clean: + lprocfs_exp_cleanup(exp); + return rc; } /* Add client data to the FILTER. We use a bitmap to locate a free space @@ -269,20 +329,20 @@ static int filter_client_add(struct obd_device *obd, struct obd_export *exp, * there's no need for extra complication here */ if (new_client) { - cl_idx = find_first_zero_bit(bitmap, LR_MAX_CLIENTS); + cl_idx = cfs_find_first_zero_bit(bitmap, LR_MAX_CLIENTS); repeat: if (cl_idx >= LR_MAX_CLIENTS) { CERROR("no room for %u client - fix LR_MAX_CLIENTS\n", cl_idx); RETURN(-EOVERFLOW); } - if (test_and_set_bit(cl_idx, bitmap)) { - cl_idx = find_next_zero_bit(bitmap, LR_MAX_CLIENTS, - cl_idx); + if (cfs_test_and_set_bit(cl_idx, bitmap)) { + cl_idx = cfs_find_next_zero_bit(bitmap, LR_MAX_CLIENTS, + cl_idx); goto repeat; } } else { - if (test_and_set_bit(cl_idx, bitmap)) { + if (cfs_test_and_set_bit(cl_idx, bitmap)) { CERROR("FILTER client %d: bit already set in bitmap!\n", cl_idx); LBUG(); @@ -292,6 +352,7 @@ static int filter_client_add(struct obd_device *obd, struct obd_export *exp, fed->fed_lr_idx = cl_idx; fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->lsd_client_start) + cl_idx * le16_to_cpu(filter->fo_fsd->lsd_client_size); + cfs_init_mutex(&fed->fed_lcd_lock); LASSERTF(fed->fed_lr_off > 0, "fed_lr_off = %llu\n", fed->fed_lr_off); CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n", @@ -315,12 +376,16 @@ 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); + target_client_add_cb, + class_export_cb_get(exp)); if (rc == 0) { - spin_lock(&exp->exp_lock); + cfs_spin_lock(&exp->exp_lock); exp->exp_need_sync = 1; - spin_unlock(&exp->exp_lock); + cfs_spin_unlock(&exp->exp_lock); } rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, fed->fed_lcd, @@ -341,31 +406,33 @@ 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; + struct lsd_client_data *lcd = fed->fed_lcd; int rc; loff_t off; ENTRY; - if (fed->fed_lcd == NULL) + if (lcd == NULL) RETURN(0); /* XXX if lcd_uuid were a real obd_uuid, I could use obd_uuid_equals */ - if (strcmp(fed->fed_lcd->lcd_uuid, obd->obd_uuid.uuid ) == 0) + if (strcmp(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, 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) { @@ -376,42 +443,43 @@ static int filter_client_free(struct obd_export *exp) /* Clear the bit _after_ zeroing out the client so we don't race with filter_client_add and zero out new clients.*/ - if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) { + if (!cfs_test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) { CERROR("FILTER client %u: bit already clear in bitmap!!\n", fed->fed_lr_idx); LBUG(); } - if (!(exp->exp_flags & OBD_OPT_FAILOVER)) { - memset(&zero_lcd, 0, sizeof zero_lcd); - 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)); - if (rc == 0) - /* update server's transno */ - filter_update_server_data(obd, filter->fo_rcvd_filp, - filter->fo_fsd, - !exp->exp_libclient); - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + /* Make sure the server's last_transno is up to date. + * This should be done before zeroing client slot so last_transno will + * be in server data or in client data in case of failure */ + filter_update_server_data(obd, filter->fo_rcvd_filp, filter->fo_fsd); + + cfs_mutex_down(&fed->fed_lcd_lock); + rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_lcd, + sizeof(zero_lcd), &off, 0); + fed->fed_lcd = NULL; + cfs_mutex_up(&fed->fed_lcd_lock); + 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", - fed->fed_lcd->lcd_uuid, fed->fed_lr_idx, fed->fed_lr_off, - LAST_RCVD, rc); - } + CDEBUG(rc == 0 ? D_INFO : D_ERROR, + "zero out client %s at idx %u/%llu in %s, rc %d\n", + lcd->lcd_uuid, fed->fed_lr_idx, fed->fed_lr_off, + LAST_RCVD, rc); - if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) { + if (!cfs_test_and_clear_bit(fed->fed_lr_idx, + filter->fo_last_rcvd_slots)) { CERROR("FILTER client %u: bit already clear in bitmap!!\n", fed->fed_lr_idx); LBUG(); } - - EXIT; + OBD_FREE_PTR(lcd); + RETURN(0); free: - OBD_FREE_PTR(fed->fed_lcd); + cfs_mutex_down(&fed->fed_lcd_lock); fed->fed_lcd = NULL; + cfs_mutex_up(&fed->fed_lcd_lock); + OBD_FREE_PTR(lcd); return 0; } @@ -425,7 +493,7 @@ static inline void filter_fmd_put_nolock(struct filter_export_data *fed, /* XXX when we have persistent reservations and the handle * is stored herein we need to drop it here. */ fed->fed_mod_count--; - list_del(&fmd->fmd_list); + cfs_list_del(&fmd->fmd_list); OBD_SLAB_FREE(fmd, ll_fmd_cachep, sizeof(*fmd)); } } @@ -439,9 +507,9 @@ void filter_fmd_put(struct obd_export *exp, struct filter_mod_data *fmd) return; fed = &exp->exp_filter_data; - spin_lock(&fed->fed_lock); + cfs_spin_lock(&fed->fed_lock); filter_fmd_put_nolock(fed, fmd); /* caller reference */ - spin_unlock(&fed->fed_lock); + cfs_spin_unlock(&fed->fed_lock); } /* expire entries from the end of the list if there are too many @@ -452,25 +520,25 @@ static void filter_fmd_expire_nolock(struct filter_obd *filter, { struct filter_mod_data *fmd, *tmp; - list_for_each_entry_safe(fmd, tmp, &fed->fed_mod_list, fmd_list) { + cfs_list_for_each_entry_safe(fmd, tmp, &fed->fed_mod_list, fmd_list) { if (fmd == keep) break; - if (time_before(jiffies, fmd->fmd_expire) && + if (cfs_time_before(jiffies, fmd->fmd_expire) && fed->fed_mod_count < filter->fo_fmd_max_num) break; - list_del_init(&fmd->fmd_list); + cfs_list_del_init(&fmd->fmd_list); filter_fmd_put_nolock(fed, fmd); /* list reference */ } } void filter_fmd_expire(struct obd_export *exp) { - spin_lock(&exp->exp_filter_data.fed_lock); + cfs_spin_lock(&exp->exp_filter_data.fed_lock); filter_fmd_expire_nolock(&exp->exp_obd->u.filter, &exp->exp_filter_data, NULL); - spin_unlock(&exp->exp_filter_data.fed_lock); + cfs_spin_unlock(&exp->exp_filter_data.fed_lock); } /* find specified objid, group in export fmd list. @@ -483,11 +551,11 @@ static struct filter_mod_data *filter_fmd_find_nolock(struct filter_obd *filter, LASSERT_SPIN_LOCKED(&fed->fed_lock); - list_for_each_entry_reverse(fmd, &fed->fed_mod_list, fmd_list) { + cfs_list_for_each_entry_reverse(fmd, &fed->fed_mod_list, fmd_list) { if (fmd->fmd_id == objid && fmd->fmd_gr == group) { found = fmd; - list_del(&fmd->fmd_list); - list_add_tail(&fmd->fmd_list, &fed->fed_mod_list); + cfs_list_del(&fmd->fmd_list); + cfs_list_add_tail(&fmd->fmd_list, &fed->fed_mod_list); fmd->fmd_expire = jiffies + filter->fo_fmd_max_age; break; } @@ -504,12 +572,12 @@ struct filter_mod_data *filter_fmd_find(struct obd_export *exp, { struct filter_mod_data *fmd; - spin_lock(&exp->exp_filter_data.fed_lock); + cfs_spin_lock(&exp->exp_filter_data.fed_lock); fmd = filter_fmd_find_nolock(&exp->exp_obd->u.filter, &exp->exp_filter_data, objid, group); if (fmd) fmd->fmd_refcount++; /* caller reference */ - spin_unlock(&exp->exp_filter_data.fed_lock); + cfs_spin_unlock(&exp->exp_filter_data.fed_lock); return fmd; } @@ -525,13 +593,14 @@ 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); + cfs_spin_lock(&fed->fed_lock); found = filter_fmd_find_nolock(&exp->exp_obd->u.filter,fed,objid,group); if (fmd_new) { if (found == NULL) { - list_add_tail(&fmd_new->fmd_list, &fed->fed_mod_list); + cfs_list_add_tail(&fmd_new->fmd_list, + &fed->fed_mod_list); fmd_new->fmd_id = objid; fmd_new->fmd_gr = group; fmd_new->fmd_refcount++; /* list reference */ @@ -547,7 +616,7 @@ struct filter_mod_data *filter_fmd_get(struct obd_export *exp, exp->exp_obd->u.filter.fo_fmd_max_age; } - spin_unlock(&fed->fed_lock); + cfs_spin_unlock(&fed->fed_lock); return found; } @@ -561,13 +630,13 @@ static void filter_fmd_drop(struct obd_export *exp, obd_id objid, obd_gr group) { struct filter_mod_data *found = NULL; - spin_lock(&exp->exp_filter_data.fed_lock); + cfs_spin_lock(&exp->exp_filter_data.fed_lock); found = filter_fmd_find_nolock(&exp->exp_filter_data, objid, group); if (found) { - list_del_init(&found->fmd_list); + cfs_list_del_init(&found->fmd_list); filter_fmd_put_nolock(&exp->exp_filter_data, found); } - spin_unlock(&exp->exp_filter_data.fed_lock); + cfs_spin_unlock(&exp->exp_filter_data.fed_lock); } #else #define filter_fmd_drop(exp, objid, group) @@ -579,29 +648,29 @@ static void filter_fmd_cleanup(struct obd_export *exp) struct filter_export_data *fed = &exp->exp_filter_data; struct filter_mod_data *fmd = NULL, *tmp; - spin_lock(&fed->fed_lock); - list_for_each_entry_safe(fmd, tmp, &fed->fed_mod_list, fmd_list) { - list_del_init(&fmd->fmd_list); + cfs_spin_lock(&fed->fed_lock); + cfs_list_for_each_entry_safe(fmd, tmp, &fed->fed_mod_list, fmd_list) { + cfs_list_del_init(&fmd->fmd_list); filter_fmd_put_nolock(fed, fmd); } - spin_unlock(&fed->fed_lock); + cfs_spin_unlock(&fed->fed_lock); } static int filter_init_export(struct obd_export *exp) { - spin_lock_init(&exp->exp_filter_data.fed_lock); + cfs_spin_lock_init(&exp->exp_filter_data.fed_lock); CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list); - spin_lock(&exp->exp_lock); + cfs_spin_lock(&exp->exp_lock); exp->exp_connecting = 1; - spin_unlock(&exp->exp_lock); + cfs_spin_unlock(&exp->exp_lock); return ldlm_init_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; @@ -610,7 +679,7 @@ static int filter_free_server_data(struct filter_obd *filter) /* assumes caller is already in kernel ctxt */ int filter_update_server_data(struct obd_device *obd, struct file *filp, - struct lr_server_data *fsd, int force_sync) + struct lr_server_data *fsd) { loff_t off = 0; int rc; @@ -622,8 +691,7 @@ 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); + rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off, 0); if (rc) CERROR("error writing lr_server_data: rc = %d\n", rc); @@ -666,6 +734,7 @@ 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; @@ -698,7 +767,10 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp) fsd->lsd_client_size = cpu_to_le16(LR_CLIENT_SIZE); fsd->lsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT); filter->fo_subdir_count = FILTER_SUBDIR_COUNT; - fsd->lsd_feature_incompat = cpu_to_le32(OBD_INCOMPAT_OST); + /* OBD_COMPAT_OST is set in filter_connect_internal when the + * MDS first connects and assigns the OST index number. */ + fsd->lsd_feature_incompat = cpu_to_le32(OBD_INCOMPAT_COMMON_LR| + OBD_INCOMPAT_OST); } else { rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off); if (rc) { @@ -721,6 +793,10 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp) cpu_to_le32(OBD_INCOMPAT_COMMON_LR))) fsd->lsd_last_transno = fsd->lsd_compat14; /* end COMPAT_146 */ + /* OBD_COMPAT_OST is set in filter_connect_internal when the + * MDS first connects and assigns the OST index number. */ + fsd->lsd_feature_incompat |= cpu_to_le32(OBD_INCOMPAT_COMMON_LR| + OBD_INCOMPAT_OST); } if (fsd->lsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) { @@ -737,7 +813,11 @@ 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); @@ -789,6 +869,8 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp) continue; } + check_lcd(obd->obd_name, cl_idx, lcd); + last_rcvd = le64_to_cpu(lcd->lcd_last_transno); /* These exports are cleaned up by filter_disconnect(), so they @@ -817,12 +899,16 @@ 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; - spin_lock(&exp->exp_lock); + /* VBR: set export last committed */ + exp->exp_last_committed = last_rcvd; + cfs_spin_lock(&exp->exp_lock); exp->exp_connecting = 0; exp->exp_in_recovery = 0; - spin_unlock(&exp->exp_lock); + cfs_spin_unlock(&exp->exp_lock); + cfs_spin_lock_bh(&obd->obd_processing_task_lock); obd->obd_max_recoverable_clients++; + cfs_spin_unlock_bh(&obd->obd_processing_task_lock); + lcd = NULL; class_export_put(exp); } @@ -838,22 +924,19 @@ static int filter_init_server_data(struct obd_device *obd, struct file * filp) OBD_FREE_PTR(lcd); obd->obd_last_committed = le64_to_cpu(fsd->lsd_last_transno); - - target_recovery_init(obd, ost_handle); - out: filter->fo_mount_count = mount_count + 1; fsd->lsd_mount_count = cpu_to_le64(filter->fo_mount_count); /* save it, so mount count and last_transno is current */ - rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1); + rc = filter_update_server_data(obd, filp, filter->fo_fsd); if (rc) GOTO(err_client, rc); RETURN(0); err_client: - target_recovery_fini(obd); + class_disconnect_exports(obd); err_fsd: filter_free_server_data(filter); RETURN(rc); @@ -939,7 +1022,7 @@ 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); + CDEBUG(D_INODE, "%s: previous %d, new %d\n", obd->obd_name, last_group, group); @@ -1012,7 +1095,7 @@ static int filter_read_group_internal(struct obd_device *obd, int group, GOTO(cleanup, rc); } - if (filter->fo_subdir_count) { + if (filter->fo_subdir_count && filter_group_is_mds(group)) { OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs)); if (tmp_subdirs == NULL) GOTO(cleanup, rc = -ENOMEM); @@ -1076,7 +1159,7 @@ static int filter_read_group_internal(struct obd_device *obd, int group, filter->fo_dentry_O_groups[group] = dentry; filter->fo_last_objid_files[group] = filp; - if (filter->fo_subdir_count) { + if (filter->fo_subdir_count && filter_group_is_mds(group)) { filter->fo_dentry_O_sub[group] = *tmp_subdirs; OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs)); } @@ -1105,7 +1188,7 @@ static int filter_read_group_internal(struct obd_device *obd, int group, if (new_files != NULL) OBD_FREE(new_files, len * sizeof(*new_files)); case 3: - if (filter->fo_subdir_count) { + if (filter->fo_subdir_count && filter_group_is_mds(group)) { for (i = 0; i < filter->fo_subdir_count; i++) { if (tmp_subdirs->dentry[i] != NULL) dput(tmp_subdirs->dentry[i]); @@ -1126,17 +1209,14 @@ static int filter_read_groups(struct obd_device *obd, int last_group, struct filter_obd *filter = &obd->u.filter; int old_count, group, rc = 0; - down(&filter->fo_init_lock); + cfs_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) break; } - up(&filter->fo_init_lock); + cfs_up(&filter->fo_init_lock); return rc; } @@ -1144,7 +1224,7 @@ static int filter_read_groups(struct obd_device *obd, int last_group, static int filter_prep_groups(struct obd_device *obd) { struct filter_obd *filter = &obd->u.filter; - struct dentry *dentry, *O_dentry; + struct dentry *O_dentry; struct file *filp; int last_group, rc = 0, cleanup_phase = 0; loff_t off = 0; @@ -1161,57 +1241,6 @@ static int filter_prep_groups(struct obd_device *obd) filter->fo_dentry_O = O_dentry; cleanup_phase = 1; /* O_dentry */ - /* Lookup "R" to tell if we're on an old OST FS and need to convert - * from O/R// to O/0//. This can be removed - * some time post 1.0 when all old-style OSTs have converted along - * with the init_objid hack. */ - dentry = ll_lookup_one_len("R", O_dentry, 1); - if (IS_ERR(dentry)) - GOTO(cleanup, rc = PTR_ERR(dentry)); - if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { - struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1); - ENTRY; - - CWARN("converting OST to new object layout\n"); - if (IS_ERR(O0_dentry)) { - rc = PTR_ERR(O0_dentry); - CERROR("error looking up O/0: rc %d\n", rc); - GOTO(cleanup_R, rc); - } - - if (O0_dentry->d_inode) { - CERROR("Both O/R and O/0 exist. Fix manually.\n"); - GOTO(cleanup_O0, rc = -EEXIST); - } - - LOCK_INODE_MUTEX(O_dentry->d_inode); - rc = ll_vfs_rename(O_dentry->d_inode, dentry, filter->fo_vfsmnt, - O_dentry->d_inode, O0_dentry, - filter->fo_vfsmnt); - UNLOCK_INODE_MUTEX(O_dentry->d_inode); - - if (rc) { - CERROR("error renaming O/R to O/0: rc %d\n", rc); - GOTO(cleanup_O0, rc); - } - filter->fo_fsd->lsd_feature_incompat |= - cpu_to_le32(OBD_INCOMPAT_GROUPS); - rc = filter_update_server_data(obd, filter->fo_rcvd_filp, - filter->fo_fsd, 1); - GOTO(cleanup_O0, rc); - - cleanup_O0: - f_dput(O0_dentry); - cleanup_R: - f_dput(dentry); - if (rc) - GOTO(cleanup, rc); - } else { - f_dput(dentry); - } - - cleanup_phase = 2; /* groups */ - /* we have to initialize all groups before first connections from * clients because they may send create/destroy for any group -bzzz */ filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700); @@ -1219,21 +1248,19 @@ static int filter_prep_groups(struct obd_device *obd) CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp)); GOTO(cleanup, rc = PTR_ERR(filp)); } - cleanup_phase = 3; /* filp */ + cleanup_phase = 2; /* filp */ rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off); if (rc) { CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc); GOTO(cleanup, rc); } - if (off == 0) { - last_group = FILTER_MIN_GROUPS; - } else { - LASSERT(last_group >= FILTER_MIN_GROUPS); - } + + if (off == 0) + last_group = FILTER_GROUP_MDS0; CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name, - FILTER_MIN_GROUPS, last_group); + FILTER_GROUP_MDS0, last_group); filter->fo_committed_group = last_group; rc = filter_read_groups(obd, last_group, 1); if (rc) @@ -1244,11 +1271,10 @@ static int filter_prep_groups(struct obd_device *obd) cleanup: switch (cleanup_phase) { - case 3: - filp_close(filp, 0); case 2: - filter_cleanup_groups(obd); + filp_close(filp, 0); case 1: + filter_cleanup_groups(obd); f_dput(filter->fo_dentry_O); filter->fo_dentry_O = NULL; default: @@ -1291,11 +1317,17 @@ 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); GOTO(err_filp, rc); } + + target_recovery_init(&filter->fo_lut, ost_handle); + /* open and create health check io file*/ file = filp_open(HEALTH_CHECK, O_RDWR | O_CREAT, 0644); if (IS_ERR(file)) { @@ -1304,7 +1336,7 @@ static int filter_prep(struct obd_device *obd) HEALTH_CHECK, rc); GOTO(err_server_data, rc); } - filter->fo_health_check_filp = file; + filter->fo_obt.obt_health_check_filp = file; if (!S_ISREG(file->f_dentry->d_inode->i_mode)) { CERROR("%s is not a regular file!: mode = %o\n", HEALTH_CHECK, file->f_dentry->d_inode->i_mode); @@ -1323,9 +1355,9 @@ out: return(rc); err_health_check: - if (filp_close(filter->fo_health_check_filp, 0)) + if (filp_close(filter->fo_obt.obt_health_check_filp, 0)) CERROR("can't close %s after error\n", HEALTH_CHECK); - filter->fo_health_check_filp = NULL; + filter->fo_obt.obt_health_check_filp = NULL; err_server_data: target_recovery_fini(obd); filter_free_server_data(filter); @@ -1349,11 +1381,11 @@ static void filter_post(struct obd_device *obd) push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); rc = filter_update_server_data(obd, filter->fo_rcvd_filp, - filter->fo_fsd, 0); + filter->fo_fsd); 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) @@ -1366,8 +1398,8 @@ static void filter_post(struct obd_device *obd) if (rc) CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc); - rc = filp_close(filter->fo_health_check_filp, 0); - filter->fo_health_check_filp = NULL; + rc = filp_close(filter->fo_obt.obt_health_check_filp, 0); + filter->fo_obt.obt_health_check_filp = NULL; if (rc) CERROR("error closing %s: rc = %d\n", HEALTH_CHECK, rc); @@ -1385,9 +1417,9 @@ static void filter_set_last_id(struct filter_obd *filter, LASSERT(filter->fo_fsd != NULL); LASSERT(group <= filter->fo_group_count); - spin_lock(&filter->fo_objidlock); + cfs_spin_lock(&filter->fo_objidlock); filter->fo_last_objids[group] = id; - spin_unlock(&filter->fo_objidlock); + cfs_spin_unlock(&filter->fo_objidlock); } obd_id filter_last_id(struct filter_obd *filter, obd_gr group) @@ -1395,12 +1427,12 @@ obd_id filter_last_id(struct filter_obd *filter, obd_gr group) obd_id id; LASSERT(filter->fo_fsd != NULL); LASSERT(group <= filter->fo_group_count); + LASSERT(filter->fo_last_objids != NULL); /* FIXME: object groups */ - spin_lock(&filter->fo_objidlock); + cfs_spin_lock(&filter->fo_objidlock); id = filter->fo_last_objids[group]; - spin_unlock(&filter->fo_objidlock); - + cfs_spin_unlock(&filter->fo_objidlock); return id; } @@ -1417,8 +1449,7 @@ 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) || - filter->fo_subdir_count == 0) + if (!filter_group_is_mds(group) || filter->fo_subdir_count == 0) return filter->fo_dentry_O_groups[group]; subdirs = &filter->fo_dentry_O_sub[group]; @@ -1515,9 +1546,8 @@ 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 } }; @@ -1529,15 +1559,19 @@ 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); - + lockh); + if (rc != ELDLM_OK) + lockh->cookie = 0; RETURN(rc); } +static void filter_fini_destroy(struct obd_device *obd, + struct lustre_handle *lockh) +{ + if (lustre_handle_is_used(lockh)) + 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 @@ -1637,7 +1671,7 @@ static enum interval_iter filter_intent_cb(struct interval_node *n, if (interval_high(n) <= size) return INTERVAL_ITER_STOP; - list_for_each_entry(lck, &node->li_group, l_sl_policy) { + cfs_list_for_each_entry(lck, &node->li_group, l_sl_policy) { /* Don't send glimpse ASTs to liblustre clients. * They aren't listening for them, and they do * entirely synchronous I/O anyways. */ @@ -1714,13 +1748,14 @@ static int filter_intent_policy(struct ldlm_namespace *ns, /* FIXME: we should change the policy function slightly, to not make * this list at all, since we just turn around and free it */ - while (!list_empty(&rpc_list)) { + while (!cfs_list_empty(&rpc_list)) { struct ldlm_lock *wlock = - list_entry(rpc_list.next, struct ldlm_lock, l_cp_ast); + cfs_list_entry(rpc_list.next, struct ldlm_lock, + l_cp_ast); LASSERT((lock->l_flags & LDLM_FL_AST_SENT) == 0); LASSERT(lock->l_flags & LDLM_FL_CP_REQD); lock->l_flags &= ~LDLM_FL_CP_REQD; - list_del_init(&wlock->l_cp_ast); + cfs_list_del_init(&wlock->l_cp_ast); LDLM_LOCK_RELEASE(wlock); } @@ -1783,7 +1818,7 @@ static int filter_intent_policy(struct ldlm_namespace *ns, * * Of course, this will all disappear when we switch to * taking liblustre locks on the OST. */ - ldlm_res_lvbo_update(res, NULL, 0, 1); + ldlm_res_lvbo_update(res, NULL, 1); } RETURN(ELDLM_LOCK_ABORTED); } @@ -1810,7 +1845,7 @@ static int filter_intent_policy(struct ldlm_namespace *ns, * sending ast is not handled. This can result in lost client writes. */ if (rc != 0) - ldlm_res_lvbo_update(res, NULL, 0, 1); + ldlm_res_lvbo_update(res, NULL, 1); lock_res(res); *reply_lvb = *res_lvb; @@ -1873,6 +1908,30 @@ 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); + + cfs_write_lock(&filter->fo_sptlrpc_lock); + sptlrpc_rule_set_free(&filter->fo_sptlrpc_rset); + filter->fo_sptlrpc_rset = tmp_rset; + cfs_write_unlock(&filter->fo_sptlrpc_lock); + + return 0; +} + /* * pre-allocate pool of iobuf's to be used by filter_{prep,commit}rw_write(). */ @@ -1899,7 +1958,8 @@ 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; @@ -1972,7 +2032,11 @@ 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; @@ -1998,39 +2062,36 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, obd->obd_lvfs_ctxt.fs = get_ds(); obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops; - init_mutex(&filter->fo_init_lock); + cfs_init_mutex(&filter->fo_init_lock); filter->fo_committed_group = 0; - - rc = filter_prep(obd); - if (rc) - GOTO(err_ops, rc); - filter->fo_destroys_in_progress = 0; for (i = 0; i < 32; i++) - sema_init(&filter->fo_create_locks[i], 1); + cfs_sema_init(&filter->fo_create_locks[i], 1); - spin_lock_init(&filter->fo_translock); - spin_lock_init(&filter->fo_objidlock); + cfs_spin_lock_init(&filter->fo_translock); + cfs_spin_lock_init(&filter->fo_objidlock); CFS_INIT_LIST_HEAD(&filter->fo_export_list); - sema_init(&filter->fo_alloc_lock, 1); + cfs_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; + rc = filter_prep(obd); + if (rc) + GOTO(err_ops, rc); + CFS_INIT_LIST_HEAD(&filter->fo_llog_list); - spin_lock_init(&filter->fo_llog_list_lock); + cfs_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, @@ -2044,12 +2105,17 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL, "filter_ldlm_cb_client", &obd->obd_ldlm_client); - rc = obd_llog_init(obd, &obd->obd_olg, obd, 1, NULL, NULL); + rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL); if (rc) { CERROR("failed to setup llogging subsystems\n"); GOTO(err_post, rc); } + cfs_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); @@ -2076,17 +2142,14 @@ 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), @@ -2205,10 +2268,18 @@ 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; + } + + ctxt = llog_group_get_ctxt(olg, LLOG_CONFIG_ORIG_CTXT); + if (ctxt) { rc2 = llog_cleanup(ctxt); - if (!rc) - rc = rc2; + if (!rc) + rc = rc2; + } RETURN(rc); } @@ -2261,13 +2332,18 @@ 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); @@ -2282,8 +2358,7 @@ cleanup_lcm: static int filter_llog_init(struct obd_device *obd, struct obd_llog_group *olg, - struct obd_device *tgt, int count, struct llog_catid *catid, - struct obd_uuid *uuid) + struct obd_device *tgt, int *index) { struct filter_obd *filter = &obd->u.filter; struct llog_ctxt *ctxt; @@ -2306,7 +2381,7 @@ 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); } @@ -2318,23 +2393,33 @@ 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. + */ + cfs_mutex_down(&ctxt->loc_sem); + if (ctxt->loc_imp) { + class_import_put(ctxt->loc_imp); + ctxt->loc_imp = NULL; + } + cfs_mutex_up(&ctxt->loc_sem); + llog_ctxt_put(ctxt); } - mutex_up(&ctxt->loc_sem); - llog_ctxt_put(ctxt); if (filter->fo_lcm) { + cfs_mutex_down(&ctxt->loc_sem); llog_recov_thread_fini(filter->fo_lcm, obd->obd_force); filter->fo_lcm = NULL; + cfs_mutex_up(&ctxt->loc_sem); } RETURN(filter_olg_fini(&obd->obd_olg)); } @@ -2348,7 +2433,7 @@ filter_find_olg_internal(struct filter_obd *filter, int group) struct obd_llog_group *olg; LASSERT_SPIN_LOCKED(&filter->fo_llog_list_lock); - list_for_each_entry(olg, &filter->fo_llog_list, olg_list) { + cfs_list_for_each_entry(olg, &filter->fo_llog_list, olg_list) { if (olg->olg_group == group) RETURN(olg); } @@ -2368,9 +2453,9 @@ struct obd_llog_group *filter_find_olg(struct obd_device *obd, int group) if (group == FILTER_GROUP_LLOG) RETURN(&obd->obd_olg); - spin_lock(&filter->fo_llog_list_lock); + cfs_spin_lock(&filter->fo_llog_list_lock); olg = filter_find_olg_internal(filter, group); - spin_unlock(&filter->fo_llog_list_lock); + cfs_spin_unlock(&filter->fo_llog_list_lock); RETURN(olg); } @@ -2390,7 +2475,7 @@ struct obd_llog_group *filter_find_create_olg(struct obd_device *obd, int group) if (group == FILTER_GROUP_LLOG) RETURN(&obd->obd_olg); - spin_lock(&filter->fo_llog_list_lock); + cfs_spin_lock(&filter->fo_llog_list_lock); olg = filter_find_olg_internal(filter, group); if (olg) { if (olg->olg_initializing) { @@ -2404,28 +2489,28 @@ struct obd_llog_group *filter_find_create_olg(struct obd_device *obd, int group) GOTO(out_unlock, olg = ERR_PTR(-ENOMEM)); llog_group_init(olg, group); - list_add(&olg->olg_list, &filter->fo_llog_list); + cfs_list_add(&olg->olg_list, &filter->fo_llog_list); olg->olg_initializing = 1; - spin_unlock(&filter->fo_llog_list_lock); + cfs_spin_unlock(&filter->fo_llog_list_lock); - rc = llog_cat_initialize(obd, olg, 1, NULL); + rc = obd_llog_init(obd, olg, obd, NULL); if (rc) { - spin_lock(&filter->fo_llog_list_lock); - list_del(&olg->olg_list); - spin_unlock(&filter->fo_llog_list_lock); + cfs_spin_lock(&filter->fo_llog_list_lock); + cfs_list_del(&olg->olg_list); + cfs_spin_unlock(&filter->fo_llog_list_lock); OBD_FREE_PTR(olg); GOTO(out, olg = ERR_PTR(-ENOMEM)); } - spin_lock(&filter->fo_llog_list_lock); + cfs_spin_lock(&filter->fo_llog_list_lock); olg->olg_initializing = 0; - spin_unlock(&filter->fo_llog_list_lock); + cfs_spin_unlock(&filter->fo_llog_list_lock); CDEBUG(D_OTHER, "%s: new llog group %u (0x%p)\n", obd->obd_name, group, olg); out: RETURN(olg); out_unlock: - spin_unlock(&filter->fo_llog_list_lock); + cfs_spin_unlock(&filter->fo_llog_list_lock); GOTO(out, olg); } @@ -2438,7 +2523,7 @@ static int filter_llog_connect(struct obd_export *exp, int rc; ENTRY; - CDEBUG(D_OTHER, "%s: LLog connect for: "LPX64"/"LPX64":%x\n", + CDEBUG(D_OTHER, "%s: LLog connect for: "LPX64"/"LPX64":%x\n", obd->obd_name, body->lgdc_logid.lgl_oid, body->lgdc_logid.lgl_ogr, body->lgdc_logid.lgl_ogen); @@ -2455,9 +2540,12 @@ static int filter_llog_connect(struct obd_export *exp, body->lgdc_ctxt_idx); CWARN("%s: Recovery from log "LPX64"/"LPX64":%x\n", - obd->obd_name, body->lgdc_logid.lgl_oid, + obd->obd_name, body->lgdc_logid.lgl_oid, body->lgdc_logid.lgl_ogr, body->lgdc_logid.lgl_ogen); + cfs_spin_lock_bh(&obd->obd_processing_task_lock); + obd->u.filter.fo_mds_ost_sync = 1; + cfs_spin_unlock_bh(&obd->obd_processing_task_lock); rc = llog_connect(ctxt, &body->lgdc_logid, &body->lgdc_gen, NULL); llog_ctxt_put(ctxt); @@ -2472,7 +2560,7 @@ static int filter_llog_preclean(struct obd_device *obd) { struct obd_llog_group *olg, *tmp; struct filter_obd *filter; - struct list_head remove_list; + cfs_list_t remove_list; int rc = 0; ENTRY; @@ -2483,17 +2571,17 @@ static int filter_llog_preclean(struct obd_device *obd) filter = &obd->u.filter; CFS_INIT_LIST_HEAD(&remove_list); - spin_lock(&filter->fo_llog_list_lock); - while (!list_empty(&filter->fo_llog_list)) { - olg = list_entry(filter->fo_llog_list.next, - struct obd_llog_group, olg_list); - list_del(&olg->olg_list); - list_add(&olg->olg_list, &remove_list); + cfs_spin_lock(&filter->fo_llog_list_lock); + while (!cfs_list_empty(&filter->fo_llog_list)) { + olg = cfs_list_entry(filter->fo_llog_list.next, + struct obd_llog_group, olg_list); + cfs_list_del(&olg->olg_list); + cfs_list_add(&olg->olg_list, &remove_list); } - spin_unlock(&filter->fo_llog_list_lock); + cfs_spin_unlock(&filter->fo_llog_list_lock); - list_for_each_entry_safe(olg, tmp, &remove_list, olg_list) { - list_del_init(&olg->olg_list); + cfs_list_for_each_entry_safe(olg, tmp, &remove_list, olg_list) { + cfs_list_del_init(&olg->olg_list); rc = filter_olg_fini(olg); if (rc) CERROR("failed to cleanup llogging subsystem for %u\n", @@ -2514,7 +2602,8 @@ 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; } @@ -2530,14 +2619,8 @@ 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); @@ -2545,10 +2628,6 @@ 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; @@ -2575,7 +2654,8 @@ static int filter_cleanup(struct obd_device *obd) } static int filter_connect_internal(struct obd_export *exp, - struct obd_connect_data *data) + struct obd_connect_data *data, + int reconnect) { struct filter_export_data *fed = &exp->exp_filter_data; @@ -2602,20 +2682,28 @@ 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); + cfs_spin_lock(&exp->exp_obd->obd_osfs_lock); left = filter_grant_space_left(exp); want = data->ocd_grant; - filter_grant(exp, fed->fed_grant, want, left); + filter_grant(exp, fed->fed_grant, want, left, (reconnect == 0)); data->ocd_grant = fed->fed_grant; - spin_unlock(&exp->exp_obd->obd_osfs_lock); + cfs_spin_unlock(&exp->exp_obd->obd_osfs_lock); CDEBUG(D_CACHE, "%s: cli %s/%p ocd_grant: %d want: " 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) { @@ -2628,8 +2716,10 @@ static int filter_connect_internal(struct obd_export *exp, /* this will only happen on the first connect */ lsd->lsd_ost_index = cpu_to_le32(data->ocd_index); lsd->lsd_feature_compat |= cpu_to_le32(OBD_COMPAT_OST); + /* sync is not needed here as filter_client_add will + * set exp_need_sync flag */ filter_update_server_data(exp->exp_obd, - filter->fo_rcvd_filp, lsd, 1); + filter->fo_rcvd_filp, lsd); } else if (index != data->ocd_index) { LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index" " %u doesn't match actual OST index" @@ -2639,6 +2729,10 @@ 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)) { @@ -2673,18 +2767,14 @@ 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); } static int filter_reconnect(const struct lu_env *env, struct obd_export *exp, struct obd_device *obd, struct obd_uuid *cluuid, - struct obd_connect_data *data) + struct obd_connect_data *data, + void *localdata) { int rc; ENTRY; @@ -2692,41 +2782,44 @@ static int filter_reconnect(const struct lu_env *env, if (exp == NULL || obd == NULL || cluuid == NULL) RETURN(-EINVAL); - rc = filter_connect_internal(exp, data); + rc = filter_connect_internal(exp, data, 1); + if (rc == 0) + filter_export_stats_init(obd, exp, localdata); RETURN(rc); } /* 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, 0); 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) { @@ -2736,17 +2829,15 @@ 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); @@ -2764,9 +2855,11 @@ cleanup: OBD_FREE_PTR(lcd); fed->fed_lcd = NULL; } - class_disconnect(exp); + class_disconnect(lexp); + lprocfs_exp_cleanup(lexp); + *exp = NULL; } else { - class_export_put(exp); + *exp = lexp; } RETURN(rc); @@ -2783,7 +2876,7 @@ static void filter_grant_sanity_check(struct obd_device *obd, const char *func) obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0; obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted; - if (list_empty(&obd->obd_exports)) + if (cfs_list_empty(&obd->obd_exports)) return; /* We don't want to do this for large machines that do lots of @@ -2791,9 +2884,9 @@ static void filter_grant_sanity_check(struct obd_device *obd, const char *func) if (obd->obd_num_exports > 100) return; - spin_lock(&obd->obd_osfs_lock); - spin_lock(&obd->obd_dev_lock); - list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) { + cfs_spin_lock(&obd->obd_osfs_lock); + cfs_spin_lock(&obd->obd_dev_lock); + cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) { int error = 0; fed = &exp->exp_filter_data; if (fed->fed_grant < 0 || fed->fed_pending < 0 || @@ -2824,8 +2917,8 @@ static void filter_grant_sanity_check(struct obd_device *obd, const char *func) fo_tot_granted = obd->u.filter.fo_tot_granted; fo_tot_pending = obd->u.filter.fo_tot_pending; fo_tot_dirty = obd->u.filter.fo_tot_dirty; - spin_unlock(&obd->obd_dev_lock); - spin_unlock(&obd->obd_osfs_lock); + cfs_spin_unlock(&obd->obd_dev_lock); + cfs_spin_unlock(&obd->obd_osfs_lock); /* Do these assertions outside the spinlocks so we don't kill system */ if (tot_granted != fo_tot_granted) @@ -2858,11 +2951,7 @@ static void filter_grant_discard(struct obd_export *exp) struct filter_obd *filter = &obd->u.filter; struct filter_export_data *fed = &exp->exp_filter_data; - spin_lock(&obd->obd_osfs_lock); - spin_lock(&obd->obd_dev_lock); - list_del_init(&exp->exp_obd_chain); - spin_unlock(&obd->obd_dev_lock); - + cfs_spin_lock(&obd->obd_osfs_lock); LASSERTF(filter->fo_tot_granted >= fed->fed_grant, "%s: tot_granted "LPU64" cli %s/%p fed_grant %ld\n", obd->obd_name, filter->fo_tot_granted, @@ -2881,7 +2970,7 @@ static void filter_grant_discard(struct obd_export *exp) fed->fed_dirty = 0; fed->fed_grant = 0; - spin_unlock(&obd->obd_osfs_lock); + cfs_spin_unlock(&obd->obd_osfs_lock); } static int filter_destroy_export(struct obd_export *exp) @@ -2893,9 +2982,7 @@ 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); @@ -2904,14 +2991,18 @@ static int filter_destroy_export(struct obd_export *exp) RETURN(0); - if (exp->exp_obd->obd_replayable) - filter_client_free(exp); - else + if (!exp->exp_obd->obd_replayable) fsfilt_sync(exp->exp_obd, exp->exp_obd->u.obt.obt_sb); 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__); @@ -2922,7 +3013,7 @@ 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; @@ -2935,8 +3026,8 @@ static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp) /* look for group with min. number, but > worked */ olg_min = NULL; group = 1 << 30; - spin_lock(&filter->fo_llog_list_lock); - list_for_each_entry(olg, &filter->fo_llog_list, olg_list) { + cfs_spin_lock(&filter->fo_llog_list_lock); + cfs_list_for_each_entry(olg, &filter->fo_llog_list, olg_list) { if (olg->olg_group <= worked) { /* this group is already synced */ continue; @@ -2949,7 +3040,7 @@ static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp) olg_min = olg; group = olg->olg_group; } - spin_unlock(&filter->fo_llog_list_lock); + cfs_spin_unlock(&filter->fo_llog_list_lock); if (olg_min == NULL) break; @@ -2959,18 +3050,20 @@ 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; @@ -2984,17 +3077,18 @@ static int filter_disconnect(struct obd_export *exp) filter_grant_sanity_check(obd, __FUNCTION__); filter_grant_discard(exp); - /* Disconnect early so that clients can't keep using export */ - rc = class_disconnect(exp); - if (exp->exp_obd->obd_namespace != NULL) - ldlm_cancel_locks_for_export(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); - fsfilt_sync(obd, obd->u.obt.obt_sb); + rc = server_disconnect_export(exp); - lprocfs_exp_cleanup(exp); + if (exp->exp_obd->obd_replayable) + filter_client_free(exp); + else + fsfilt_sync(obd, obd->u.obt.obt_sb); - /* flush any remaining cancel messages out to the target */ - filter_sync_llogs(obd, exp); class_export_put(exp); RETURN(rc); } @@ -3016,7 +3110,6 @@ static void filter_revimp_update(struct obd_export *exp) static int filter_ping(struct obd_export *exp) { filter_fmd_expire(exp); - return 0; } @@ -3055,7 +3148,8 @@ static int filter_getattr(struct obd_export *exp, struct obd_info *oinfo) int rc = 0; ENTRY; - rc = filter_auth_capa(exp, NULL, oinfo_mdsno(oinfo), + LASSERT(oinfo->oi_oa->o_valid & OBD_MD_FLGROUP); + rc = filter_auth_capa(exp, NULL, oinfo->oi_oa->o_gr, oinfo_capa(oinfo), CAPA_OPC_META_READ); if (rc) RETURN(rc); @@ -3122,9 +3216,11 @@ 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 page *page = NULL; struct iattr iattr; void *handle; ENTRY; @@ -3144,11 +3240,31 @@ 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); + + /* Let's pin the last page so that ldiskfs_truncate + * should not start GFP_FS allocation. */ + if (ia_valid & ATTR_SIZE) { + page = grab_cache_page(inode->i_mapping, + iattr.ia_size >> PAGE_CACHE_SHIFT); + if (page == NULL) + GOTO(out_unlock, rc = -ENOMEM); + + unlock_page(page); } /* If the inode still has SUID+SGID bits set (see filter_precreate()) @@ -3195,8 +3311,8 @@ int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry, GOTO(out_unlock, rc = PTR_ERR(handle)); } if (oa->o_valid & OBD_MD_FLFLAGS) { - rc = fsfilt_iocontrol(exp->exp_obd, inode, NULL, - EXT3_IOC_SETFLAGS, (long)&oa->o_flags); + rc = fsfilt_iocontrol(exp->exp_obd, dentry, + FSFILT_IOC_SETFLAGS, (long)&oa->o_flags); } else { rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1); if (fcc != NULL) @@ -3213,7 +3329,7 @@ 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; @@ -3228,18 +3344,26 @@ 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) - UNLOCK_INODE_MUTEX(inode); + if (page) + page_cache_release(page); + 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)); @@ -3259,60 +3383,70 @@ 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; + + LASSERT(oa->o_valid & OBD_MD_FLGROUP); + rc = filter_auth_capa(exp, NULL, oa->o_gr, capa, opc); if (rc) RETURN(rc); + if (oa->o_valid & (OBD_MD_FLUID | OBD_MD_FLGID)) { + rc = filter_capa_fixoa(exp, oa, oa->o_gr, 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)); filter = &exp->exp_obd->u.filter; 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); @@ -3321,20 +3455,19 @@ int filter_setattr(struct obd_export *exp, struct obd_info *oinfo, if (res != NULL) { LDLM_RESOURCE_ADDREF(res); - rc = ldlm_res_lvbo_update(res, NULL, 0, 0); + rc = ldlm_res_lvbo_update(res, NULL, 0); LDLM_RESOURCE_DELREF(res); 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; out_unlock: - unlock_kernel(); f_dput(dentry); pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL); return rc; @@ -3398,15 +3531,16 @@ 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); @@ -3416,7 +3550,7 @@ static int filter_destroy_precreated(struct obd_export *exp, struct obdo *oa, doa.o_gr = oa->o_gr; doa.o_mode = S_IFREG; - if (!test_bit(doa.o_gr, &filter->fo_destroys_in_progress)) { + if (!cfs_test_bit(doa.o_gr, &filter->fo_destroys_in_progress)) { CERROR("%s:["LPU64"] destroys_in_progress already cleared\n", exp->exp_obd->obd_name, doa.o_gr); RETURN(0); @@ -3424,27 +3558,39 @@ 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); - clear_bit(doa.o_gr, &filter->fo_destroys_in_progress); + 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; + } + cfs_clear_bit(doa.o_gr, &filter->fo_destroys_in_progress); RETURN(rc); } @@ -3462,28 +3608,30 @@ 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); RETURN(0); } /* This causes inflight precreates to abort and drop lock */ - set_bit(group, &filter->fo_destroys_in_progress); - down(&filter->fo_create_locks[group]); - if (!test_bit(group, &filter->fo_destroys_in_progress)) { + cfs_set_bit(group, &filter->fo_destroys_in_progress); + cfs_down(&filter->fo_create_locks[group]); + if (!cfs_test_bit(group, &filter->fo_destroys_in_progress)) { CERROR("%s:["LPU64"] destroys_in_progress already cleared\n", exp->exp_obd->obd_name, group); - up(&filter->fo_create_locks[group]); + cfs_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); } @@ -3495,17 +3643,17 @@ static int filter_handle_precreate(struct obd_export *exp, struct obdo *oa, GOTO(out, rc); } else { /* XXX: Used by MDS for the first time! */ - clear_bit(group, &filter->fo_destroys_in_progress); + cfs_clear_bit(group, &filter->fo_destroys_in_progress); } } else { - down(&filter->fo_create_locks[group]); + cfs_down(&filter->fo_create_locks[group]); if (oti->oti_conn_cnt < exp->exp_conn_cnt) { CERROR("%s: dropping old precreate request\n", 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 (!filter_group_is_mds(group) || oa->o_id == 0) diff = 1; else diff = oa->o_id - filter_last_id(filter, group); @@ -3527,7 +3675,7 @@ static int filter_handle_precreate(struct obd_export *exp, struct obdo *oa, /* else diff == 0 */ GOTO(out, rc = 0); out: - up(&filter->fo_create_locks[group]); + cfs_up(&filter->fo_create_locks[group]); return rc; } @@ -3542,10 +3690,10 @@ static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs, /* at least try to account for cached pages. its still racey and * might be under-reporting if clients haven't announced their * caches with brw recently */ - spin_lock(&obd->obd_osfs_lock); + cfs_spin_lock(&obd->obd_osfs_lock); rc = fsfilt_statfs(obd, obd->u.obt.obt_sb, max_age); memcpy(osfs, &obd->obd_osfs, sizeof(*osfs)); - spin_unlock(&obd->obd_osfs_lock); + cfs_spin_unlock(&obd->obd_osfs_lock); CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64 " pending "LPU64" free "LPU64" avail "LPU64"\n", @@ -3559,11 +3707,27 @@ static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs, ((filter->fo_tot_dirty + filter->fo_tot_pending + osfs->os_bsize - 1) >> blockbits)); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC)) { + struct lr_server_data *lsd = filter->fo_fsd; + int index = le32_to_cpu(lsd->lsd_ost_index); + + if (obd_fail_val == -1 || + index == obd_fail_val) + osfs->os_bfree = osfs->os_bavail = 2; + else if (obd_fail_loc & OBD_FAIL_ONCE) + obd_fail_loc &= ~OBD_FAILED; /* reset flag */ + } + /* set EROFS to state field if FS is mounted as RDONLY. The goal is to * 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; + osfs->os_state = 0; + + if (filter->fo_obt.obt_sb->s_flags & MS_RDONLY) + osfs->os_state = OS_STATE_READONLY; + + if (filter->fo_raid_degraded) + osfs->os_state |= OS_STATE_DEGRADED; RETURN(rc); } @@ -3591,6 +3755,19 @@ 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; + + cfs_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; + cfs_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. @@ -3611,6 +3788,7 @@ 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; @@ -3628,7 +3806,8 @@ static int filter_precreate(struct obd_device *obd, struct obdo *oa, OBD_ALLOC(osfs, sizeof(*osfs)); if (osfs == NULL) RETURN(-ENOMEM); - rc = filter_statfs(obd, osfs, cfs_time_current_64() - HZ, 0); + rc = filter_statfs(obd, osfs, cfs_time_current_64() - CFS_HZ, + 0); if (rc == 0 && osfs->os_bavail < (osfs->os_blocks >> 10)) { CDEBUG(D_RPCTRACE,"%s: not enough space for create " LPU64"\n", obd->obd_name, osfs->os_bavail << @@ -3647,7 +3826,7 @@ static int filter_precreate(struct obd_device *obd, struct obdo *oa, for (i = 0; i < *num && err == 0; i++) { int cleanup_phase = 0; - if (test_bit(group, &filter->fo_destroys_in_progress)) { + if (cfs_test_bit(group, &filter->fo_destroys_in_progress)) { CWARN("%s: create aborted by destroy\n", obd->obd_name); rc = -EAGAIN; @@ -3713,7 +3892,7 @@ static int filter_precreate(struct obd_device *obd, struct obdo *oa, cleanup_phase = 3; CDEBUG(D_INODE, "%s: filter_precreate(od->o_gr="LPU64 - ",od->o_id="LPU64")\n", obd->obd_name, group, + ",od->o_id="LPU64")\n", obd->obd_name, group, next_id); /* We mark object SUID+SGID to flag it for accepting UID+GID @@ -3723,9 +3902,19 @@ 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); @@ -3754,7 +3943,7 @@ set_last_id: if (rc) break; - if (time_after(jiffies, enough_time)) { + if (cfs_time_after(jiffies, enough_time)) { CDEBUG(D_RPCTRACE, "%s: precreate slow - want %d got %d \n", obd->obd_name, *num, i); @@ -3784,7 +3973,7 @@ 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); @@ -3823,9 +4012,9 @@ static int filter_create(struct obd_export *exp, struct obdo *oa, rc = -EINVAL; } else { diff = 1; - down(&filter->fo_create_locks[oa->o_gr]); + cfs_down(&filter->fo_create_locks[oa->o_gr]); rc = filter_precreate(obd, oa, oa->o_gr, &diff); - up(&filter->fo_create_locks[oa->o_gr]); + cfs_up(&filter->fo_create_locks[oa->o_gr]); } } else { rc = filter_handle_precreate(exp, oa, oa->o_gr, oti); @@ -3848,12 +4037,13 @@ 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; @@ -3862,6 +4052,10 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, ENTRY; LASSERT(oa->o_valid & OBD_MD_FLGROUP); + rc = filter_auth_capa(exp, NULL, oa->o_gr, + (struct lustre_capa *)capa, CAPA_OPC_OSS_DESTROY); + if (rc) + RETURN(rc); obd = exp->exp_obd; filter = &obd->u.filter; @@ -3884,15 +4078,14 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, if (oa->o_valid & OBD_MD_FLCOOKIE) { struct llog_ctxt *ctxt; struct obd_llog_group *olg; - fcc = &oa->o_lcookie; + olg = filter_find_olg(obd, oa->o_gr); if (!olg) { CERROR(" %s: can not find olg of group %d\n", obd->obd_name, (int)oa->o_gr); GOTO(cleanup, rc = PTR_ERR(olg)); } - llog_group_set_export(olg, exp); - + fcc = &oa->o_lcookie; ctxt = llog_group_get_ctxt(olg, fcc->lgc_subsys + 1); llog_cancel(ctxt, NULL, 1, fcc, 0); llog_ctxt_put(ctxt); @@ -3901,7 +4094,9 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, GOTO(cleanup, rc = -ENOENT); } - filter_prepare_destroy(obd, oa->o_id, oa->o_gr); + rc = filter_prepare_destroy(obd, oa->o_id, oa->o_gr, &lockh); + if (rc) + GOTO(cleanup, rc); /* Our MDC connection is established by the MDS to us */ if (oa->o_valid & OBD_MD_FLCOOKIE) { @@ -3917,12 +4112,28 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, * down(i_zombie) down(i_zombie) * restart transaction * (see BUG 4180) -bzzz + * + * take i_alloc_sem too to prevent other threads from writing to the + * file while we are truncating it. This can cause lock ordering issue + * between page lock, i_mutex & starting new journal handle. + * (see bug 20321) -johann */ + down_write(&dchild->d_inode->i_alloc_sem); LOCK_INODE_MUTEX(dchild->d_inode); + + /* VBR: version recovery check */ + rc = filter_version_get_check(exp, oti, dchild->d_inode); + if (rc) { + UNLOCK_INODE_MUTEX(dchild->d_inode); + up_write(&dchild->d_inode->i_alloc_sem); + GOTO(cleanup, rc); + } + handle = fsfilt_start_log(obd, dchild->d_inode, FSFILT_OP_SETATTR, NULL, 1); if (IS_ERR(handle)) { UNLOCK_INODE_MUTEX(dchild->d_inode); + up_write(&dchild->d_inode->i_alloc_sem); GOTO(cleanup, rc = PTR_ERR(handle)); } @@ -3931,6 +4142,7 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1); rc2 = fsfilt_commit(obd, dchild->d_inode, handle, 0); UNLOCK_INODE_MUTEX(dchild->d_inode); + up_write(&dchild->d_inode->i_alloc_sem); if (rc) GOTO(cleanup, rc); if (rc2) @@ -3975,7 +4187,7 @@ 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; @@ -3991,6 +4203,8 @@ 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)); @@ -4007,9 +4221,8 @@ 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; } @@ -4031,13 +4244,10 @@ 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); } @@ -4051,17 +4261,18 @@ static int filter_sync(struct obd_export *exp, struct obdo *oa, int rc, rc2; ENTRY; - rc = filter_auth_capa(exp, NULL, obdo_mdsno(oa), + LASSERT(oa->o_valid & OBD_MD_FLGROUP); + rc = filter_auth_capa(exp, NULL, oa->o_gr, (struct lustre_capa *)capa, CAPA_OPC_OSS_WRITE); if (rc) RETURN(rc); 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); } @@ -4165,30 +4376,94 @@ static int filter_get_info(struct obd_export *exp, __u32 keylen, memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap)); 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); - } + rc = fsfilt_iocontrol(obd, dentry, FSFILT_IOC_FIEMAP, + (long)fiemap); 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 */ + cfs_spin_lock(&exp->exp_obd->obd_osfs_lock); + filter_grant_incoming(exp, &body->oa); + cfs_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; @@ -4198,6 +4473,7 @@ 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); @@ -4206,36 +4482,22 @@ 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, @@ -4247,7 +4509,7 @@ 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); } @@ -4321,8 +4583,8 @@ static int filter_health_check(struct obd_device *obd) rc = 1; #ifdef USE_HEALTH_CHECK_WRITE - LASSERT(filter->fo_health_check_filp != NULL); - rc |= !!lvfs_check_io_health(obd, filter->fo_health_check_filp); + LASSERT(filter->fo_obt.obt_health_check_filp != NULL); + rc |= !!lvfs_check_io_health(obd, filter->fo_obt.obt_health_check_filp); #endif return rc; } @@ -4341,39 +4603,13 @@ 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; } @@ -4422,11 +4658,15 @@ extern quota_interface_t filter_quota_interface; static int __init obdfilter_init(void) { struct lprocfs_static_vars lvars; - int rc; + int rc, i; + + /** sanity check for group<->mdsno conversion */ + for (i = 0; i < 32; i++) + LASSERT(objgrp_to_mdsno(mdt_to_obd_objgrp(i)) == i); lprocfs_filter_init_vars(&lvars); - request_module("lquota"); + cfs_request_module("%s", "lquota"); OBD_ALLOC(obdfilter_created_scratchpad, OBDFILTER_CREATED_SCRATCHPAD_ENTRIES * sizeof(*obdfilter_created_scratchpad));