int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd, const struct lu_fid *pfid, const char *name, __u32 mode, struct lu_fid *fid) { struct dt_object *parent, *dto; int rc; ENTRY; LASSERT(!fid_is_zero(pfid)); parent = dt_locate(env, mdd->mdd_bottom, pfid); if (unlikely(IS_ERR(parent))) RETURN(PTR_ERR(parent)); /* create local file/dir, if @fid is passed then try to use it */ if (fid_is_zero(fid)) dto = local_file_find_or_create(env, mdd->mdd_los, parent, name, mode); else dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom, fid, parent, name, mode); if (IS_ERR(dto)) GOTO(out_put, rc = PTR_ERR(dto)); *fid = *lu_object_fid(&dto->do_lu); /* since stack is not fully set up the local_storage uses own stack * and we should drop its object from cache */ lu_object_put_nocache(env, &dto->do_lu); EXIT; out_put: lu_object_put(env, &parent->do_lu); return 0; }
/** * This is helper function to get llog directory object. It is used by named * llog operations to find/insert/delete llog entry from llog directory. * * \param[in] env execution environment * \param[in] ctxt llog context * * \retval dt_object of llog directory * \retval ERR_PTR of negative value on error */ struct dt_object *llog_osd_dir_get(const struct lu_env *env, struct llog_ctxt *ctxt) { struct dt_device *dt; struct dt_thread_info *dti = dt_info(env); struct dt_object *dir; int rc; dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt; if (ctxt->loc_dir == NULL) { rc = dt_root_get(env, dt, &dti->dti_fid); if (rc) return ERR_PTR(rc); dir = dt_locate(env, dt, &dti->dti_fid); if (!IS_ERR(dir) && !dt_try_as_dir(env, dir)) { lu_object_put(env, &dir->do_lu); return ERR_PTR(-ENOTDIR); } } else { lu_object_get(&ctxt->loc_dir->do_lu); dir = ctxt->loc_dir; } return dir; }
/** * open the PENDING directory for device \a mdd * * The PENDING directory persistently tracks files and directories that were * unlinked from the namespace (nlink == 0) but are still held open by clients. * Those inodes shouldn't be deleted if the MDS crashes, because the clients * would not be able to recover and reopen those files. Instead, these inodes * are linked into the PENDING directory on disk, and only deleted if all * clients close them, or the MDS finishes client recovery without any client * reopening them (i.e. former clients didn't join recovery). * \param d mdd device being started. * * \retval 0 success * \retval -ve index operation error. * */ int orph_index_init(const struct lu_env *env, struct mdd_device *mdd) { struct lu_fid fid; struct dt_object *d; int rc = 0; ENTRY; /* create PENDING dir */ fid_zero(&fid); rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid, orph_index_name, S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO, &fid); if (rc < 0) RETURN(rc); d = dt_locate(env, mdd->mdd_child, &fid); if (IS_ERR(d)) RETURN(PTR_ERR(d)); LASSERT(lu_object_exists(&d->do_lu)); if (!dt_try_as_dir(env, d)) { CERROR("%s: \"%s\" is not an index: rc = %d\n", mdd2obd_dev(mdd)->obd_name, orph_index_name, rc); lu_object_put(env, &d->do_lu); RETURN(-ENOTDIR); } mdd->mdd_orphans = d; RETURN(0); }
/** * Write the special file which contains the list of llog catalogs IDs * * This function writes the CATALOG file which contains the array of llog * catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT * number. * * \param[in] env execution environment * \param[in] d corresponding storage device * \param[in] idx position to start from, usually OST/MDT index * \param[in] count how many catalog IDs to write * \param[out] idarray the buffer with the data to write. * \param[in] fid LLOG_CATALOGS_OID for CATALOG object * * \retval 0 on successful write of catalog IDs * \retval negative value on error */ int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d, int idx, int count, struct llog_catid *idarray, const struct lu_fid *fid) { struct llog_thread_info *lgi = llog_info(env); struct dt_object *o = NULL; struct thandle *th; int rc, size; if (count == 0) RETURN(0); LASSERT(d); size = sizeof(*idarray) * count; lgi->lgi_off = idx * sizeof(*idarray); lgi->lgi_fid = *fid; o = dt_locate(env, d, &lgi->lgi_fid); if (IS_ERR(o)) RETURN(PTR_ERR(o)); if (!dt_object_exists(o)) GOTO(out, rc = -ENOENT); rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA); if (rc) GOTO(out, rc); if (!S_ISREG(lgi->lgi_attr.la_mode)) { CERROR("%s: CATALOGS is not a regular file!: mode = %o\n", o->do_lu.lo_dev->ld_obd->obd_name, lgi->lgi_attr.la_mode); GOTO(out, rc = -ENOENT); } th = dt_trans_create(env, d); if (IS_ERR(th)) GOTO(out, rc = PTR_ERR(th)); lgi->lgi_buf.lb_len = size; lgi->lgi_buf.lb_buf = idarray; rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th); if (rc) GOTO(out, rc); rc = dt_trans_start_local(env, d, th); if (rc) GOTO(out_trans, rc); rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th); if (rc) CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n", idx, rc); out_trans: dt_trans_stop(env, d, th); out: lu_object_put(env, &o->do_lu); RETURN(rc); }
/** * Look-up accounting object to collect space usage information for user * or group. * * \param env - is the environment passed by the caller * \param dev - is the dt_device storing the accounting object * \param type - is the quota type, either USRQUOTA or GRPQUOTA */ struct dt_object *acct_obj_lookup(const struct lu_env *env, struct dt_device *dev, int type) { struct lquota_thread_info *qti = lquota_info(env); struct dt_object *obj = NULL; ENTRY; lu_local_obj_fid(&qti->qti_fid, qtype2acct_oid(type)); /* lookup the accounting object */ obj = dt_locate(env, dev, &qti->qti_fid); if (IS_ERR(obj)) RETURN(obj); if (!dt_object_exists(obj)) { dt_object_put(env, obj); RETURN(ERR_PTR(-ENOENT)); } if (obj->do_index_ops == NULL) { int rc; /* set up indexing operations */ rc = obj->do_ops->do_index_try(env, obj, &dt_acct_features); if (rc) { CERROR("%s: failed to set up indexing operations for %s" " acct object rc:%d\n", dev->dd_lu_dev.ld_obd->obd_name, qtype_name(type), rc); dt_object_put(env, obj); RETURN(ERR_PTR(rc)); } } RETURN(obj); }
int lfsck_bookmark_setup(const struct lu_env *env, struct lfsck_instance *lfsck) { struct dt_object *root; struct dt_object *obj; int rc; ENTRY; root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid); if (IS_ERR(root)) RETURN(PTR_ERR(root)); dt_try_as_dir(env, root); obj = local_file_find_or_create(env, lfsck->li_los, root, lfsck_bookmark_name, S_IFREG | S_IRUGO | S_IWUSR); lu_object_put(env, &root->do_lu); if (IS_ERR(obj)) RETURN(PTR_ERR(obj)); lfsck->li_bookmark_obj = obj; rc = lfsck_bookmark_load(env, lfsck); if (rc == -ENODATA) rc = lfsck_bookmark_init(env, lfsck); RETURN(rc); }
/* * Helper function looking up & creating if not found an index file with a * dynamic fid. */ static struct dt_object * lquota_disk_find_create(const struct lu_env *env, struct dt_device *dev, struct dt_object *parent, struct lu_fid *fid, const struct dt_index_features *idx_feat, char *name) { struct lquota_thread_info *qti = lquota_info(env); struct dt_object *obj; struct local_oid_storage *los; int rc; ENTRY; /* Set up local storage */ rc = local_oid_storage_init(env, dev, fid, &los); if (rc) RETURN(ERR_PTR(rc)); /* lookup/create slave index file */ obj = local_index_find_or_create(env, los, parent, name, LQUOTA_MODE, idx_feat); if (IS_ERR(obj)) GOTO(out, obj); /* local_oid_storage_fini() will finalize the local storage device, * we have to open the object in another device stack */ qti->qti_fid = obj->do_lu.lo_header->loh_fid; lu_object_put_nocache(env, &obj->do_lu); obj = dt_locate(env, dev, &qti->qti_fid); if (IS_ERR(obj)) GOTO(out, obj); out: local_oid_storage_fini(env, los); RETURN(obj); }
/* * Look-up a slave index file. * * \param env - is the environment passed by the caller * \param dev - is the backend dt_device where to look-up/create the slave index * \param parent - is the parent directory where to lookup the slave index * \param glb_fid - is the fid of the global index file associated with this * slave index. * \param uuid - is the uuid of slave which is (re)connecting to the master * target * * \retval - pointer to the dt_object of the slave index on success, * appropriate error on failure */ struct dt_object *lquota_disk_slv_find(const struct lu_env *env, struct dt_device *dev, struct dt_object *parent, const struct lu_fid *glb_fid, struct obd_uuid *uuid) { struct lquota_thread_info *qti = lquota_info(env); struct dt_object *slv_idx; int rc; ENTRY; LASSERT(uuid != NULL); CDEBUG(D_QUOTA, "lookup slave index file for %s\n", obd_uuid2str(uuid)); /* generate filename associated with the slave */ rc = lquota_disk_slv_filename(glb_fid, uuid, qti->qti_buf); if (rc) RETURN(ERR_PTR(rc)); /* lookup slave index file */ rc = dt_lookup_dir(env, parent, qti->qti_buf, &qti->qti_fid); if (rc) RETURN(ERR_PTR(rc)); /* name is found, get the object */ slv_idx = dt_locate(env, dev, &qti->qti_fid); if (IS_ERR(slv_idx)) RETURN(slv_idx); if (slv_idx->do_index_ops == NULL) { rc = slv_idx->do_ops->do_index_try(env, slv_idx, &dt_quota_slv_features); if (rc) { CERROR("%s: failed to setup slave index operations for " "%s, rc:%d\n", dev->dd_lu_dev.ld_obd->obd_name, obd_uuid2str(uuid), rc); lu_object_put(env, &slv_idx->do_lu); slv_idx = ERR_PTR(rc); } } RETURN(slv_idx); }
int lfsck_bookmark_setup(const struct lu_env *env, struct lfsck_instance *lfsck) { struct dt_object *root; struct dt_object *obj; int rc; ENTRY; root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid); if (IS_ERR(root)) RETURN(PTR_ERR(root)); if (unlikely(!dt_try_as_dir(env, root))) { lfsck_object_put(env, root); RETURN(-ENOTDIR); } obj = local_file_find_or_create(env, lfsck->li_los, root, LFSCK_BOOKMARK, S_IFREG | S_IRUGO | S_IWUSR); lfsck_object_put(env, root); if (IS_ERR(obj)) RETURN(PTR_ERR(obj)); lfsck->li_bookmark_obj = obj; rc = lfsck_bookmark_load(env, lfsck); if (rc == 0) { struct lfsck_bookmark *mb = &lfsck->li_bookmark_ram; /* It is upgraded from old release, set it as * LFSCK_ASYNC_WIN_DEFAULT to avoid memory pressure. */ if (unlikely(mb->lb_async_windows == 0)) { mb->lb_async_windows = LFSCK_ASYNC_WIN_DEFAULT; mutex_lock(&lfsck->li_mutex); rc = lfsck_bookmark_store(env, lfsck); mutex_unlock(&lfsck->li_mutex); } } else if (rc == -ENODATA) { rc = lfsck_bookmark_init(env, lfsck); } RETURN(rc); }
/** * Initialize slave index object to collect local quota limit for user or group. * * \param env - is the environment passed by the caller * \param dev - is the dt_device storing the slave index object * \param type - is the quota type, either USRQUOTA or GRPQUOTA */ static struct dt_object *quota_obj_lookup(const struct lu_env *env, struct dt_device *dev, int type) { struct lquota_thread_info *qti = lquota_info(env); struct dt_object *obj = NULL; ENTRY; qti->qti_fid.f_seq = FID_SEQ_QUOTA; qti->qti_fid.f_oid = qtype2slv_oid(type); qti->qti_fid.f_ver = 0; /* lookup the quota object */ obj = dt_locate(env, dev, &qti->qti_fid); if (IS_ERR(obj)) RETURN(obj); if (!dt_object_exists(obj)) { dt_object_put(env, obj); RETURN(ERR_PTR(-ENOENT)); } if (obj->do_index_ops == NULL) { int rc; /* set up indexing operations */ rc = obj->do_ops->do_index_try(env, obj, &dt_quota_slv_features); if (rc) { CERROR("%s: failed to set up indexing operations for %s" " slave index object rc:%d\n", dev->dd_lu_dev.ld_obd->obd_name, qtype_name(type), rc); dt_object_put(env, obj); RETURN(ERR_PTR(rc)); } } RETURN(obj); }
int lfsck_namespace_setup(const struct lu_env *env, struct lfsck_instance *lfsck) { struct lfsck_component *com; struct lfsck_namespace *ns; struct dt_object *root = NULL; struct dt_object *obj; int rc; ENTRY; LASSERT(lfsck->li_master); OBD_ALLOC_PTR(com); if (com == NULL) RETURN(-ENOMEM); CFS_INIT_LIST_HEAD(&com->lc_link); CFS_INIT_LIST_HEAD(&com->lc_link_dir); init_rwsem(&com->lc_sem); atomic_set(&com->lc_ref, 1); com->lc_lfsck = lfsck; com->lc_type = LT_NAMESPACE; com->lc_ops = &lfsck_namespace_ops; com->lc_file_size = sizeof(struct lfsck_namespace); OBD_ALLOC(com->lc_file_ram, com->lc_file_size); if (com->lc_file_ram == NULL) GOTO(out, rc = -ENOMEM); OBD_ALLOC(com->lc_file_disk, com->lc_file_size); if (com->lc_file_disk == NULL) GOTO(out, rc = -ENOMEM); root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid); if (IS_ERR(root)) GOTO(out, rc = PTR_ERR(root)); if (unlikely(!dt_try_as_dir(env, root))) GOTO(out, rc = -ENOTDIR); obj = local_index_find_or_create(env, lfsck->li_los, root, lfsck_namespace_name, S_IFREG | S_IRUGO | S_IWUSR, &dt_lfsck_features); if (IS_ERR(obj)) GOTO(out, rc = PTR_ERR(obj)); com->lc_obj = obj; rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features); if (rc != 0) GOTO(out, rc); rc = lfsck_namespace_load(env, com); if (rc > 0) rc = lfsck_namespace_reset(env, com, true); else if (rc == -ENODATA) rc = lfsck_namespace_init(env, com); if (rc != 0) GOTO(out, rc); ns = com->lc_file_ram; switch (ns->ln_status) { case LS_INIT: case LS_COMPLETED: case LS_FAILED: case LS_STOPPED: spin_lock(&lfsck->li_lock); cfs_list_add_tail(&com->lc_link, &lfsck->li_list_idle); spin_unlock(&lfsck->li_lock); break; default: CERROR("%s: unknown lfsck_namespace status: rc = %u\n", lfsck_lfsck2name(lfsck), ns->ln_status); /* fall through */ case LS_SCANNING_PHASE1: case LS_SCANNING_PHASE2: /* No need to store the status to disk right now. * If the system crashed before the status stored, * it will be loaded back when next time. */ ns->ln_status = LS_CRASHED; /* fall through */ case LS_PAUSED: case LS_CRASHED: spin_lock(&lfsck->li_lock); cfs_list_add_tail(&com->lc_link, &lfsck->li_list_scan); cfs_list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir); spin_unlock(&lfsck->li_lock); break; } GOTO(out, rc = 0); out: if (root != NULL && !IS_ERR(root)) lu_object_put(env, &root->do_lu); if (rc != 0) lfsck_component_cleanup(env, com); return rc; }
/** * Read the special file which contains the list of llog catalogs IDs * * This function reads the CATALOGS file which contains the array of llog * catalogs IDs. The main purpose of this file is to store OSP llogs indexed * by OST/MDT number. * * \param[in] env execution environment * \param[in] d corresponding storage device * \param[in] idx position to start from, usually OST/MDT index * \param[in] count how many catalog IDs to read * \param[out] idarray the buffer for the data. If it is NULL then * function returns just number of catalog IDs * in the file. * \param[in] fid LLOG_CATALOGS_OID for CATALOG object * * \retval 0 on successful read of catalog IDs * \retval negative value on error * \retval positive value which is number of records in * the file if \a idarray is NULL */ int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d, int idx, int count, struct llog_catid *idarray, const struct lu_fid *fid) { struct llog_thread_info *lgi = llog_info(env); struct dt_object *o = NULL; struct thandle *th; int rc, size; ENTRY; LASSERT(d); size = sizeof(*idarray) * count; lgi->lgi_off = idx * sizeof(*idarray); lgi->lgi_fid = *fid; o = dt_locate(env, d, &lgi->lgi_fid); if (IS_ERR(o)) RETURN(PTR_ERR(o)); if (!dt_object_exists(o)) { th = dt_trans_create(env, d); if (IS_ERR(th)) GOTO(out, rc = PTR_ERR(th)); lgi->lgi_attr.la_valid = LA_MODE; lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR; lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG); rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL, &lgi->lgi_dof, th); if (rc) GOTO(out_trans, rc); rc = dt_trans_start_local(env, d, th); if (rc) GOTO(out_trans, rc); dt_write_lock(env, o, 0); if (!dt_object_exists(o)) rc = dt_create(env, o, &lgi->lgi_attr, NULL, &lgi->lgi_dof, th); dt_write_unlock(env, o); out_trans: dt_trans_stop(env, d, th); if (rc) GOTO(out, rc); } rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA); if (rc) GOTO(out, rc); if (!S_ISREG(lgi->lgi_attr.la_mode)) { CERROR("%s: CATALOGS is not a regular file!: mode = %o\n", o->do_lu.lo_dev->ld_obd->obd_name, lgi->lgi_attr.la_mode); GOTO(out, rc = -ENOENT); } CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n", (int)lgi->lgi_attr.la_size, size); /* return just number of llogs */ if (idarray == NULL) { rc = lgi->lgi_attr.la_size / sizeof(*idarray); GOTO(out, rc); } /* read for new ost index or for empty file */ memset(idarray, 0, size); if (lgi->lgi_attr.la_size <= lgi->lgi_off) GOTO(out, rc = 0); if (lgi->lgi_attr.la_size < lgi->lgi_off + size) size = lgi->lgi_attr.la_size - lgi->lgi_off; lgi->lgi_buf.lb_buf = idarray; lgi->lgi_buf.lb_len = size; rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off); if (rc) { CERROR("%s: error reading CATALOGS: rc = %d\n", o->do_lu.lo_dev->ld_obd->obd_name, rc); GOTO(out, rc); } EXIT; out: lu_object_put(env, &o->do_lu); RETURN(rc); }
/** * Object updates between Targets. Because all the updates has been * dis-assemblied into object updates at sender side, so OUT will * call OSD API directly to execute these updates. * * In DNE phase I all of the updates in the request need to be executed * in one transaction, and the transaction has to be synchronously. * * Please refer to lustre/include/lustre/lustre_idl.h for req/reply * format. */ int out_handle(struct tgt_session_info *tsi) { const struct lu_env *env = tsi->tsi_env; struct tgt_thread_info *tti = tgt_th_info(env); struct thandle_exec_args *ta = &tti->tti_tea; struct req_capsule *pill = tsi->tsi_pill; struct dt_device *dt = tsi->tsi_tgt->lut_bottom; struct update_buf *ubuf; struct update *update; struct update_reply *update_reply; int bufsize; int count; int old_batchid = -1; unsigned off; int i; int rc = 0; int rc1 = 0; ENTRY; req_capsule_set(pill, &RQF_UPDATE_OBJ); bufsize = req_capsule_get_size(pill, &RMF_UPDATE, RCL_CLIENT); if (bufsize != UPDATE_BUFFER_SIZE) { CERROR("%s: invalid bufsize %d: rc = %d\n", tgt_name(tsi->tsi_tgt), bufsize, -EPROTO); RETURN(err_serious(-EPROTO)); } ubuf = req_capsule_client_get(pill, &RMF_UPDATE); if (ubuf == NULL) { CERROR("%s: No buf!: rc = %d\n", tgt_name(tsi->tsi_tgt), -EPROTO); RETURN(err_serious(-EPROTO)); } if (le32_to_cpu(ubuf->ub_magic) != UPDATE_BUFFER_MAGIC) { CERROR("%s: invalid magic %x expect %x: rc = %d\n", tgt_name(tsi->tsi_tgt), le32_to_cpu(ubuf->ub_magic), UPDATE_BUFFER_MAGIC, -EPROTO); RETURN(err_serious(-EPROTO)); } count = le32_to_cpu(ubuf->ub_count); if (count <= 0) { CERROR("%s: No update!: rc = %d\n", tgt_name(tsi->tsi_tgt), -EPROTO); RETURN(err_serious(-EPROTO)); } req_capsule_set_size(pill, &RMF_UPDATE_REPLY, RCL_SERVER, UPDATE_BUFFER_SIZE); rc = req_capsule_server_pack(pill); if (rc != 0) { CERROR("%s: Can't pack response: rc = %d\n", tgt_name(tsi->tsi_tgt), rc); RETURN(rc); } /* Prepare the update reply buffer */ update_reply = req_capsule_server_get(pill, &RMF_UPDATE_REPLY); if (update_reply == NULL) RETURN(err_serious(-EPROTO)); update_init_reply_buf(update_reply, count); tti->tti_u.update.tti_update_reply = update_reply; rc = out_tx_start(env, dt, ta); if (rc != 0) RETURN(rc); tti->tti_mult_trans = !req_is_replay(tgt_ses_req(tsi)); /* Walk through updates in the request to execute them synchronously */ off = cfs_size_round(offsetof(struct update_buf, ub_bufs[0])); for (i = 0; i < count; i++) { struct tgt_handler *h; struct dt_object *dt_obj; update = (struct update *)((char *)ubuf + off); if (old_batchid == -1) { old_batchid = update->u_batchid; } else if (old_batchid != update->u_batchid) { /* Stop the current update transaction, * create a new one */ rc = out_tx_end(env, ta); if (rc != 0) RETURN(rc); rc = out_tx_start(env, dt, ta); if (rc != 0) RETURN(rc); old_batchid = update->u_batchid; } fid_le_to_cpu(&update->u_fid, &update->u_fid); if (!fid_is_sane(&update->u_fid)) { CERROR("%s: invalid FID "DFID": rc = %d\n", tgt_name(tsi->tsi_tgt), PFID(&update->u_fid), -EPROTO); GOTO(out, rc = err_serious(-EPROTO)); } dt_obj = dt_locate(env, dt, &update->u_fid); if (IS_ERR(dt_obj)) GOTO(out, rc = PTR_ERR(dt_obj)); tti->tti_u.update.tti_dt_object = dt_obj; tti->tti_u.update.tti_update = update; tti->tti_u.update.tti_update_reply_index = i; h = out_handler_find(update->u_type); if (likely(h != NULL)) { /* For real modification RPC, check if the update * has been executed */ if (h->th_flags & MUTABOR) { struct ptlrpc_request *req = tgt_ses_req(tsi); if (out_check_resent(env, dt, dt_obj, req, out_reconstruct, update_reply, i)) GOTO(next, rc); } rc = h->th_act(tsi); } else { CERROR("%s: The unsupported opc: 0x%x\n", tgt_name(tsi->tsi_tgt), update->u_type); lu_object_put(env, &dt_obj->do_lu); GOTO(out, rc = -ENOTSUPP); } next: lu_object_put(env, &dt_obj->do_lu); if (rc < 0) GOTO(out, rc); off += cfs_size_round(update_size(update)); } out: rc1 = out_tx_end(env, ta); if (rc == 0) rc = rc1; RETURN(rc); }
/** * Object updates between Targets. Because all the updates has been * dis-assemblied into object updates at sender side, so OUT will * call OSD API directly to execute these updates. * * In DNE phase I all of the updates in the request need to be executed * in one transaction, and the transaction has to be synchronously. * * Please refer to lustre/include/lustre/lustre_idl.h for req/reply * format. */ int out_handle(struct tgt_session_info *tsi) { const struct lu_env *env = tsi->tsi_env; struct tgt_thread_info *tti = tgt_th_info(env); struct thandle_exec_args *ta = &tti->tti_tea; struct req_capsule *pill = tsi->tsi_pill; struct dt_device *dt = tsi->tsi_tgt->lut_bottom; struct out_update_header *ouh; struct out_update_buffer *oub = NULL; struct object_update *update; struct object_update_reply *reply; struct ptlrpc_bulk_desc *desc = NULL; struct l_wait_info lwi; void **update_bufs; int current_batchid = -1; __u32 update_buf_count; unsigned int i; unsigned int reply_index = 0; int rc = 0; int rc1 = 0; int ouh_size, reply_size; int updates; ENTRY; req_capsule_set(pill, &RQF_OUT_UPDATE); ouh_size = req_capsule_get_size(pill, &RMF_OUT_UPDATE_HEADER, RCL_CLIENT); if (ouh_size <= 0) RETURN(err_serious(-EPROTO)); ouh = req_capsule_client_get(pill, &RMF_OUT_UPDATE_HEADER); if (ouh == NULL) RETURN(err_serious(-EPROTO)); if (ouh->ouh_magic != OUT_UPDATE_HEADER_MAGIC) { CERROR("%s: invalid update buffer magic %x expect %x: " "rc = %d\n", tgt_name(tsi->tsi_tgt), ouh->ouh_magic, UPDATE_REQUEST_MAGIC, -EPROTO); RETURN(err_serious(-EPROTO)); } update_buf_count = ouh->ouh_count; if (update_buf_count == 0) RETURN(err_serious(-EPROTO)); OBD_ALLOC(update_bufs, sizeof(*update_bufs) * update_buf_count); if (update_bufs == NULL) RETURN(err_serious(-ENOMEM)); if (ouh->ouh_inline_length > 0) { update_bufs[0] = ouh->ouh_inline_data; } else { struct out_update_buffer *tmp; oub = req_capsule_client_get(pill, &RMF_OUT_UPDATE_BUF); if (oub == NULL) GOTO(out_free, rc = err_serious(-EPROTO)); desc = ptlrpc_prep_bulk_exp(pill->rc_req, update_buf_count, PTLRPC_BULK_OPS_COUNT, PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KVEC, MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops); if (desc == NULL) GOTO(out_free, rc = err_serious(-ENOMEM)); tmp = oub; for (i = 0; i < update_buf_count; i++, tmp++) { if (tmp->oub_size >= OUT_MAXREQSIZE) GOTO(out_free, rc = err_serious(-EPROTO)); OBD_ALLOC(update_bufs[i], tmp->oub_size); if (update_bufs[i] == NULL) GOTO(out_free, rc = err_serious(-ENOMEM)); desc->bd_frag_ops->add_iov_frag(desc, update_bufs[i], tmp->oub_size); } pill->rc_req->rq_bulk_write = 1; rc = sptlrpc_svc_prep_bulk(pill->rc_req, desc); if (rc != 0) GOTO(out_free, rc = err_serious(rc)); rc = target_bulk_io(pill->rc_req->rq_export, desc, &lwi); if (rc < 0) GOTO(out_free, rc = err_serious(rc)); } /* validate the request and calculate the total update count and * set it to reply */ reply_size = 0; updates = 0; for (i = 0; i < update_buf_count; i++) { struct object_update_request *our; int j; our = update_bufs[i]; if (ptlrpc_req_need_swab(pill->rc_req)) lustre_swab_object_update_request(our); if (our->ourq_magic != UPDATE_REQUEST_MAGIC) { CERROR("%s: invalid update buffer magic %x" " expect %x: rc = %d\n", tgt_name(tsi->tsi_tgt), our->ourq_magic, UPDATE_REQUEST_MAGIC, -EPROTO); GOTO(out_free, rc = err_serious(-EPROTO)); } updates += our->ourq_count; /* need to calculate reply size */ for (j = 0; j < our->ourq_count; j++) { update = object_update_request_get(our, j, NULL); if (update == NULL) GOTO(out, rc = err_serious(-EPROTO)); if (ptlrpc_req_need_swab(pill->rc_req)) lustre_swab_object_update(update); if (!fid_is_sane(&update->ou_fid)) { CERROR("%s: invalid FID "DFID": rc = %d\n", tgt_name(tsi->tsi_tgt), PFID(&update->ou_fid), -EPROTO); GOTO(out, rc = err_serious(-EPROTO)); } /* XXX: what ou_result_size can be considered safe? */ reply_size += sizeof(reply->ourp_lens[0]); reply_size += sizeof(struct object_update_result); reply_size += update->ou_result_size; } } reply_size += sizeof(*reply); if (unlikely(reply_size > ouh->ouh_reply_size)) { CERROR("%s: too small reply buf %u for %u, need %u at least\n", tgt_name(tsi->tsi_tgt), ouh->ouh_reply_size, updates, reply_size); GOTO(out_free, rc = err_serious(-EPROTO)); } req_capsule_set_size(pill, &RMF_OUT_UPDATE_REPLY, RCL_SERVER, ouh->ouh_reply_size); rc = req_capsule_server_pack(pill); if (rc != 0) { CERROR("%s: Can't pack response: rc = %d\n", tgt_name(tsi->tsi_tgt), rc); GOTO(out_free, rc = err_serious(-EPROTO)); } /* Prepare the update reply buffer */ reply = req_capsule_server_get(pill, &RMF_OUT_UPDATE_REPLY); if (reply == NULL) GOTO(out_free, rc = -EPROTO); reply->ourp_magic = UPDATE_REPLY_MAGIC; reply->ourp_count = updates; tti->tti_u.update.tti_update_reply = reply; tti->tti_mult_trans = !req_is_replay(tgt_ses_req(tsi)); /* Walk through updates in the request to execute them */ for (i = 0; i < update_buf_count; i++) { struct tgt_handler *h; struct dt_object *dt_obj; int update_count; struct object_update_request *our; int j; our = update_bufs[i]; update_count = our->ourq_count; for (j = 0; j < update_count; j++) { update = object_update_request_get(our, j, NULL); dt_obj = dt_locate(env, dt, &update->ou_fid); if (IS_ERR(dt_obj)) GOTO(out, rc = PTR_ERR(dt_obj)); if (dt->dd_record_fid_accessed) { lfsck_pack_rfa(&tti->tti_lr, lu_object_fid(&dt_obj->do_lu), LE_FID_ACCESSED, LFSCK_TYPE_LAYOUT); tgt_lfsck_in_notify(env, dt, &tti->tti_lr, NULL); } tti->tti_u.update.tti_dt_object = dt_obj; tti->tti_u.update.tti_update = update; tti->tti_u.update.tti_update_reply_index = reply_index; h = out_handler_find(update->ou_type); if (unlikely(h == NULL)) { CERROR("%s: unsupported opc: 0x%x\n", tgt_name(tsi->tsi_tgt), update->ou_type); GOTO(next, rc = -ENOTSUPP); } /* Check resend case only for modifying RPC */ if (h->th_flags & MUTABOR) { struct ptlrpc_request *req = tgt_ses_req(tsi); if (out_check_resent(env, dt, dt_obj, req, out_reconstruct, reply, reply_index)) GOTO(next, rc = 0); } /* start transaction for modification RPC only */ if (h->th_flags & MUTABOR && current_batchid == -1) { current_batchid = update->ou_batchid; rc = out_tx_start(env, dt, ta, tsi->tsi_exp); if (rc != 0) GOTO(next, rc); if (update->ou_flags & UPDATE_FL_SYNC) ta->ta_handle->th_sync = 1; } /* Stop the current update transaction, if the update * has different batchid, or read-only update */ if (((current_batchid != update->ou_batchid) || !(h->th_flags & MUTABOR)) && ta->ta_handle != NULL) { rc = out_tx_end(env, ta, rc); current_batchid = -1; if (rc != 0) GOTO(next, rc); /* start a new transaction if needed */ if (h->th_flags & MUTABOR) { rc = out_tx_start(env, dt, ta, tsi->tsi_exp); if (rc != 0) GOTO(next, rc); if (update->ou_flags & UPDATE_FL_SYNC) ta->ta_handle->th_sync = 1; current_batchid = update->ou_batchid; } } rc = h->th_act(tsi); next: reply_index++; lu_object_put(env, &dt_obj->do_lu); if (rc < 0) GOTO(out, rc); } } out: if (current_batchid != -1) { rc1 = out_tx_end(env, ta, rc); if (rc == 0) rc = rc1; } out_free: if (update_bufs != NULL) { if (oub != NULL) { for (i = 0; i < update_buf_count; i++, oub++) { if (update_bufs[i] != NULL) OBD_FREE(update_bufs[i], oub->oub_size); } } OBD_FREE(update_bufs, sizeof(*update_bufs) * update_buf_count); } if (desc != NULL) ptlrpc_free_bulk(desc); RETURN(rc); }
static int lfsck_namespace_reset(const struct lu_env *env, struct lfsck_component *com, bool init) { struct lfsck_instance *lfsck = com->lc_lfsck; struct lfsck_namespace *ns = com->lc_file_ram; struct dt_object *root; struct dt_object *dto; int rc; ENTRY; root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid); if (IS_ERR(root)) RETURN(PTR_ERR(root)); if (unlikely(!dt_try_as_dir(env, root))) { lu_object_put(env, &root->do_lu); RETURN(-ENOTDIR); } down_write(&com->lc_sem); if (init) { memset(ns, 0, sizeof(*ns)); } else { __u32 count = ns->ln_success_count; __u64 last_time = ns->ln_time_last_complete; memset(ns, 0, sizeof(*ns)); ns->ln_success_count = count; ns->ln_time_last_complete = last_time; } ns->ln_magic = LFSCK_NAMESPACE_MAGIC; ns->ln_status = LS_INIT; rc = local_object_unlink(env, lfsck->li_bottom, root, lfsck_namespace_name); if (rc != 0) GOTO(out, rc); lfsck_object_put(env, com->lc_obj); com->lc_obj = NULL; dto = local_index_find_or_create(env, lfsck->li_los, root, lfsck_namespace_name, S_IFREG | S_IRUGO | S_IWUSR, &dt_lfsck_features); if (IS_ERR(dto)) GOTO(out, rc = PTR_ERR(dto)); com->lc_obj = dto; rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features); if (rc != 0) GOTO(out, rc); rc = lfsck_namespace_store(env, com, true); GOTO(out, rc); out: up_write(&com->lc_sem); lu_object_put(env, &root->do_lu); return rc; }
/* * Set up quota directory (either "quota_master" or "quota_slave") for a QMT or * QSD instance. This function is also used to create per-pool directory on * the quota master. * The directory is created with a local sequence if it does not exist already. * This function is called at ->ldo_prepare time when the full device stack is * configured. * * \param env - is the environment passed by the caller * \param dev - is the dt_device where to create the quota directory * \param parent - is the parent directory. If not specified, the directory * will be created under the root directory * \param name - is the name of quota directory to be created * * \retval - pointer to quota root dt_object on success, appropriate error * on failure */ struct dt_object *lquota_disk_dir_find_create(const struct lu_env *env, struct dt_device *dev, struct dt_object *parent, const char *name) { struct lquota_thread_info *qti = lquota_info(env); struct dt_object *qt_dir = NULL; struct local_oid_storage *los = NULL; int rc; ENTRY; /* Set up local storage to create the quota directory. * We use the sequence reserved for local named objects */ lu_local_name_obj_fid(&qti->qti_fid, 1); rc = local_oid_storage_init(env, dev, &qti->qti_fid, &los); if (rc) RETURN(ERR_PTR(rc)); if (parent == NULL) { /* Fetch dt object associated with root directory */ rc = dt_root_get(env, dev, &qti->qti_fid); if (rc) GOTO(out, rc); parent = dt_locate_at(env, dev, &qti->qti_fid, dev->dd_lu_dev.ld_site->ls_top_dev, NULL); if (IS_ERR(parent)) GOTO(out, rc = PTR_ERR(parent)); } else { lu_object_get(&parent->do_lu); } /* create quota directory to be used for all quota index files */ qt_dir = local_file_find_or_create(env, los, parent, name, S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO); if (IS_ERR(qt_dir)) GOTO(out, rc = PTR_ERR(qt_dir)); /* local_oid_storage_fini() will finalize the local storage device, * we have to open the object in another device stack */ qti->qti_fid = qt_dir->do_lu.lo_header->loh_fid; lu_object_put_nocache(env, &qt_dir->do_lu); qt_dir = dt_locate(env, dev, &qti->qti_fid); if (IS_ERR(qt_dir)) GOTO(out, rc = PTR_ERR(qt_dir)); if (!dt_try_as_dir(env, qt_dir)) GOTO(out, rc = -ENOTDIR); EXIT; out: if (parent != NULL && !IS_ERR(parent)) lu_object_put(env, &parent->do_lu); if (los != NULL) local_oid_storage_fini(env, los); if (rc) { if (qt_dir != NULL && !IS_ERR(qt_dir)) lu_object_put(env, &qt_dir->do_lu); qt_dir = ERR_PTR(rc); } return qt_dir; }
/** * Object updates between Targets. Because all the updates has been * dis-assemblied into object updates at sender side, so OUT will * call OSD API directly to execute these updates. * * In DNE phase I all of the updates in the request need to be executed * in one transaction, and the transaction has to be synchronously. * * Please refer to lustre/include/lustre/lustre_idl.h for req/reply * format. */ int out_handle(struct tgt_session_info *tsi) { const struct lu_env *env = tsi->tsi_env; struct tgt_thread_info *tti = tgt_th_info(env); struct thandle_exec_args *ta = &tti->tti_tea; struct req_capsule *pill = tsi->tsi_pill; struct dt_device *dt = tsi->tsi_tgt->lut_bottom; struct object_update_request *ureq; struct object_update *update; struct object_update_reply *reply; int bufsize; int count; int old_batchid = -1; int i; int rc = 0; int rc1 = 0; ENTRY; req_capsule_set(pill, &RQF_OUT_UPDATE); ureq = req_capsule_client_get(pill, &RMF_OUT_UPDATE); if (ureq == NULL) { CERROR("%s: No buf!: rc = %d\n", tgt_name(tsi->tsi_tgt), -EPROTO); RETURN(err_serious(-EPROTO)); } bufsize = req_capsule_get_size(pill, &RMF_OUT_UPDATE, RCL_CLIENT); if (bufsize != object_update_request_size(ureq)) { CERROR("%s: invalid bufsize %d: rc = %d\n", tgt_name(tsi->tsi_tgt), bufsize, -EPROTO); RETURN(err_serious(-EPROTO)); } if (ureq->ourq_magic != UPDATE_REQUEST_MAGIC) { CERROR("%s: invalid update buffer magic %x expect %x: " "rc = %d\n", tgt_name(tsi->tsi_tgt), ureq->ourq_magic, UPDATE_REQUEST_MAGIC, -EPROTO); RETURN(err_serious(-EPROTO)); } count = ureq->ourq_count; if (count <= 0) { CERROR("%s: empty update: rc = %d\n", tgt_name(tsi->tsi_tgt), -EPROTO); RETURN(err_serious(-EPROTO)); } req_capsule_set_size(pill, &RMF_OUT_UPDATE_REPLY, RCL_SERVER, OUT_UPDATE_REPLY_SIZE); rc = req_capsule_server_pack(pill); if (rc != 0) { CERROR("%s: Can't pack response: rc = %d\n", tgt_name(tsi->tsi_tgt), rc); RETURN(rc); } /* Prepare the update reply buffer */ reply = req_capsule_server_get(pill, &RMF_OUT_UPDATE_REPLY); if (reply == NULL) RETURN(err_serious(-EPROTO)); object_update_reply_init(reply, count); tti->tti_u.update.tti_update_reply = reply; rc = out_tx_start(env, dt, ta, tsi->tsi_exp); if (rc != 0) RETURN(rc); tti->tti_mult_trans = !req_is_replay(tgt_ses_req(tsi)); /* Walk through updates in the request to execute them synchronously */ for (i = 0; i < count; i++) { struct tgt_handler *h; struct dt_object *dt_obj; update = object_update_request_get(ureq, i, NULL); if (update == NULL) GOTO(out, rc = -EPROTO); if (ptlrpc_req_need_swab(pill->rc_req)) lustre_swab_object_update(update); if (old_batchid == -1) { old_batchid = update->ou_batchid; } else if (old_batchid != update->ou_batchid) { /* Stop the current update transaction, * create a new one */ rc = out_tx_end(env, ta); if (rc < 0) RETURN(rc); rc = out_tx_start(env, dt, ta, tsi->tsi_exp); if (rc != 0) RETURN(rc); old_batchid = update->ou_batchid; } if (!fid_is_sane(&update->ou_fid)) { CERROR("%s: invalid FID "DFID": rc = %d\n", tgt_name(tsi->tsi_tgt), PFID(&update->ou_fid), -EPROTO); GOTO(out, rc = err_serious(-EPROTO)); } dt_obj = dt_locate(env, dt, &update->ou_fid); if (IS_ERR(dt_obj)) GOTO(out, rc = PTR_ERR(dt_obj)); if (dt->dd_record_fid_accessed) { lfsck_pack_rfa(&tti->tti_lr, lu_object_fid(&dt_obj->do_lu)); tgt_lfsck_in_notify(env, dt, &tti->tti_lr); } tti->tti_u.update.tti_dt_object = dt_obj; tti->tti_u.update.tti_update = update; tti->tti_u.update.tti_update_reply_index = i; h = out_handler_find(update->ou_type); if (likely(h != NULL)) { /* For real modification RPC, check if the update * has been executed */ if (h->th_flags & MUTABOR) { struct ptlrpc_request *req = tgt_ses_req(tsi); if (out_check_resent(env, dt, dt_obj, req, out_reconstruct, reply, i)) GOTO(next, rc); } rc = h->th_act(tsi); } else { CERROR("%s: The unsupported opc: 0x%x\n", tgt_name(tsi->tsi_tgt), update->ou_type); lu_object_put(env, &dt_obj->do_lu); GOTO(out, rc = -ENOTSUPP); } next: lu_object_put(env, &dt_obj->do_lu); if (rc < 0) GOTO(out, rc); } out: rc1 = out_tx_end(env, ta); if (rc == 0) rc = rc1; RETURN(rc); }
int fld_index_init(const struct lu_env *env, struct lu_server_fld *fld, struct dt_device *dt, int type) { struct dt_object *dt_obj = NULL; struct lu_fid fid; struct lu_attr *attr = NULL; struct lu_seq_range *range = NULL; struct fld_thread_info *info; struct dt_object_format dof; struct dt_it *it; const struct dt_it_ops *iops; int rc; __u32 index; ENTRY; info = lu_context_key_get(&env->le_ctx, &fld_thread_key); LASSERT(info != NULL); lu_local_obj_fid(&fid, FLD_INDEX_OID); OBD_ALLOC_PTR(attr); if (attr == NULL) RETURN(-ENOMEM); memset(attr, 0, sizeof(*attr)); attr->la_valid = LA_MODE; attr->la_mode = S_IFREG | 0666; dof.dof_type = DFT_INDEX; dof.u.dof_idx.di_feat = &fld_index_features; dt_obj = dt_locate(env, dt, &fid); if (IS_ERR(dt_obj)) { rc = PTR_ERR(dt_obj); dt_obj = NULL; GOTO(out, rc); } LASSERT(dt_obj != NULL); if (!dt_object_exists(dt_obj)) { lu_object_put(env, &dt_obj->do_lu); dt_obj = dt_find_or_create(env, dt, &fid, &dof, attr); fld->lsf_new = 1; if (IS_ERR(dt_obj)) { rc = PTR_ERR(dt_obj); CERROR("%s: Can't find \"%s\" obj %d\n", fld->lsf_name, fld_index_name, rc); dt_obj = NULL; GOTO(out, rc); } } fld->lsf_obj = dt_obj; rc = dt_obj->do_ops->do_index_try(env, dt_obj, &fld_index_features); if (rc != 0) { CERROR("%s: File \"%s\" is not an index: rc = %d!\n", fld->lsf_name, fld_index_name, rc); GOTO(out, rc); } range = &info->fti_rec; /* Load fld entry to cache */ iops = &dt_obj->do_index_ops->dio_it; it = iops->init(env, dt_obj, 0); if (IS_ERR(it)) GOTO(out, rc = PTR_ERR(it)); rc = iops->load(env, it, 0); if (rc < 0) GOTO(out_it_fini, rc); if (rc > 0) { /* Load FLD entry into server cache */ do { rc = iops->rec(env, it, (struct dt_rec *)range, 0); if (rc != 0) GOTO(out_it_put, rc); LASSERT(range != NULL); range_be_to_cpu(range, range); rc = fld_cache_insert(fld->lsf_cache, range); if (rc != 0) GOTO(out_it_put, rc); rc = iops->next(env, it); } while (rc == 0); } else { fld->lsf_new = 1; } rc = fld_name_to_index(fld->lsf_name, &index); if (rc < 0) GOTO(out_it_put, rc); else rc = 0; if (index == 0 && type == LU_SEQ_RANGE_MDT) { /* Note: fld_insert_entry will detect whether these * special entries already exist inside FLDB */ mutex_lock(&fld->lsf_lock); rc = fld_insert_special_entries(env, fld); mutex_unlock(&fld->lsf_lock); if (rc != 0) { CERROR("%s: insert special entries failed!: rc = %d\n", fld->lsf_name, rc); GOTO(out_it_put, rc); } } out_it_put: iops->put(env, it); out_it_fini: iops->fini(env, it); out: if (attr != NULL) OBD_FREE_PTR(attr); if (rc < 0) { if (dt_obj != NULL) lu_object_put(env, &dt_obj->do_lu); fld->lsf_obj = NULL; } RETURN(rc); }