+ obd_uuid2str(&tgt_desc->ltd_uuid), rc);
+ OBD_FREE_PTR(tgt_desc);
+ }
+ } else {
+ mutex_unlock(<d->ltd_mutex);
+ up_read(<d->ltd_rw_sem);
+ }
+}
+
+static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
+{
+ cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
+ int rc = 0;
+ ENTRY;
+
+ /* grab write reference on the lod. Relocating the array requires
+ * exclusive access */
+
+ down_write(<d->ltd_rw_sem);
+ if (newsize <= ltd->ltd_tgts_size)
+ /* someone else has already resize the array */
+ GOTO(out, rc = 0);
+
+ /* allocate new bitmap */
+ new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
+ if (!new_bitmap)
+ GOTO(out, rc = -ENOMEM);
+
+ if (ltd->ltd_tgts_size > 0) {
+ /* the bitmap already exists, we need
+ * to copy data from old one */
+ cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
+ old_bitmap = ltd->ltd_tgt_bitmap;
+ }
+
+ ltd->ltd_tgts_size = newsize;
+ ltd->ltd_tgt_bitmap = new_bitmap;
+
+ if (old_bitmap)
+ CFS_FREE_BITMAP(old_bitmap);
+
+ CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
+
+ EXIT;
+out:
+ up_write(<d->ltd_rw_sem);
+ return rc;
+}
+
+/*
+ * Connect LOD to a new OSP and add it to the device table.
+ *
+ * \param env - is the environment passed by the caller
+ * \param lod - is the LOD device to be connected to the new OSP
+ * \param osp - is the name of OSP device name about to be added
+ * \param index - is the OSP index
+ * \param gen - is the generation number
+ * \param tgt_index - is the group of the OSP.
+ * \param type - is the type of device (mdc or osc)
+ */
+int lod_add_device(const struct lu_env *env, struct lod_device *lod,
+ char *osp, unsigned index, unsigned gen, int tgt_index,
+ char *type, int active)
+{
+ struct obd_connect_data *data = NULL;
+ struct obd_export *exp = NULL;
+ struct obd_device *obd;
+ struct lu_device *ldev;
+ struct dt_device *d;
+ int rc;
+ struct lod_tgt_desc *tgt_desc;
+ struct lod_tgt_descs *ltd;
+ struct obd_uuid obd_uuid;
+ ENTRY;
+
+ CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
+
+ if (gen <= 0) {
+ CERROR("request to add OBD %s with invalid generation: %d\n",
+ osp, gen);
+ RETURN(-EINVAL);
+ }
+
+ obd_str2uuid(&obd_uuid, osp);
+
+ obd = class_find_client_obd(&obd_uuid, LUSTRE_OSP_NAME,
+ &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
+ if (obd == NULL) {
+ CERROR("can't find %s device\n", osp);
+ RETURN(-EINVAL);
+ }
+
+ OBD_ALLOC_PTR(data);
+ if (data == NULL)
+ RETURN(-ENOMEM);
+
+ data->ocd_connect_flags = OBD_CONNECT_INDEX | OBD_CONNECT_VERSION;
+ data->ocd_version = LUSTRE_VERSION_CODE;
+ data->ocd_index = index;
+
+ if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
+ data->ocd_connect_flags |= OBD_CONNECT_AT |
+ OBD_CONNECT_FULL20 |
+ OBD_CONNECT_INDEX |
+#ifdef HAVE_LRU_RESIZE_SUPPORT
+ OBD_CONNECT_LRU_RESIZE |
+#endif
+ OBD_CONNECT_MDS |
+ OBD_CONNECT_OSS_CAPA |
+ OBD_CONNECT_REQPORTAL |
+ OBD_CONNECT_SKIP_ORPHAN |
+ OBD_CONNECT_FID |
+ OBD_CONNECT_LVB_TYPE |
+ OBD_CONNECT_VERSION |
+ OBD_CONNECT_PINGLESS;
+
+ data->ocd_group = tgt_index;
+ ltd = &lod->lod_ost_descs;
+ } else {
+ struct obd_import *imp = obd->u.cli.cl_import;
+
+ data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
+ data->ocd_connect_flags |= OBD_CONNECT_ACL |
+ OBD_CONNECT_MDS_CAPA |
+ OBD_CONNECT_OSS_CAPA |
+ OBD_CONNECT_IBITS |
+ OBD_CONNECT_MDS_MDS |
+ OBD_CONNECT_FID |
+ OBD_CONNECT_AT |
+ OBD_CONNECT_FULL20 |
+ OBD_CONNECT_PINGLESS;
+ /* XXX set MDS-MDS flags, remove this when running this
+ * on client*/
+ data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
+ spin_lock(&imp->imp_lock);
+ imp->imp_server_timeout = 1;
+ spin_unlock(&imp->imp_lock);
+ imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
+ CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
+ obd->obd_name);
+ ltd = &lod->lod_mdt_descs;
+ }
+
+ rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
+ OBD_FREE_PTR(data);
+ if (rc) {
+ CERROR("%s: cannot connect to next dev %s (%d)\n",
+ obd->obd_name, osp, rc);
+ GOTO(out_free, rc);
+ }
+
+ LASSERT(obd->obd_lu_dev);
+ LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
+
+ ldev = obd->obd_lu_dev;
+ d = lu2dt_dev(ldev);
+
+ /* Allocate ost descriptor and fill it */
+ OBD_ALLOC_PTR(tgt_desc);
+ if (!tgt_desc)
+ GOTO(out_conn, rc = -ENOMEM);
+
+ tgt_desc->ltd_tgt = d;
+ tgt_desc->ltd_exp = exp;
+ tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
+ tgt_desc->ltd_gen = gen;
+ tgt_desc->ltd_index = index;
+ tgt_desc->ltd_active = active;
+
+ lod_getref(ltd);
+ if (index >= ltd->ltd_tgts_size) {
+ /* we have to increase the size of the lod_osts array */
+ __u32 newsize;
+
+ newsize = max(ltd->ltd_tgts_size, (__u32)2);
+ while (newsize < index + 1)
+ newsize = newsize << 1;
+
+ /* lod_bitmap_resize() needs lod_rw_sem
+ * which we hold with th reference */
+ lod_putref(lod, ltd);
+
+ rc = ltd_bitmap_resize(ltd, newsize);
+ if (rc)
+ GOTO(out_desc, rc);
+
+ lod_getref(ltd);
+ }
+
+ mutex_lock(<d->ltd_mutex);
+ if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
+ CERROR("%s: device %d is registered already\n", obd->obd_name,
+ index);
+ GOTO(out_mutex, rc = -EEXIST);
+ }
+
+ if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
+ OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
+ if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
+ CERROR("can't allocate index to add %s\n",
+ obd->obd_name);
+ GOTO(out_mutex, rc = -ENOMEM);
+ }
+ }
+
+ if (!strcmp(LUSTRE_OSC_NAME, type)) {
+ /* pool and qos are not supported for MDS stack yet */
+ rc = lod_ost_pool_add(&lod->lod_pool_info, index,
+ lod->lod_osts_size);
+ if (rc) {
+ CERROR("%s: can't set up pool, failed with %d\n",
+ obd->obd_name, rc);
+ GOTO(out_mutex, rc);
+ }
+
+ rc = qos_add_tgt(lod, tgt_desc);
+ if (rc) {
+ CERROR("%s: qos_add_tgt failed with %d\n",
+ obd->obd_name, rc);
+ GOTO(out_pool, rc);
+ }
+
+ /* The new OST is now a full citizen */
+ if (index >= lod->lod_desc.ld_tgt_count)
+ lod->lod_desc.ld_tgt_count = index + 1;
+ if (active)
+ lod->lod_desc.ld_active_tgt_count++;
+ }
+
+ LTD_TGT(ltd, index) = tgt_desc;
+ cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
+ ltd->ltd_tgtnr++;
+ mutex_unlock(<d->ltd_mutex);
+ lod_putref(lod, ltd);
+ if (lod->lod_recovery_completed)
+ ldev->ld_ops->ldo_recovery_complete(env, ldev);
+
+ RETURN(0);
+
+out_pool:
+ lod_ost_pool_remove(&lod->lod_pool_info, index);
+out_mutex:
+ mutex_unlock(<d->ltd_mutex);
+ lod_putref(lod, ltd);
+out_desc:
+ OBD_FREE_PTR(tgt_desc);
+out_conn:
+ obd_disconnect(exp);
+out_free:
+ return rc;
+}
+
+/*
+ * helper function to schedule OST removal from the device table
+ */
+static void __lod_del_device(struct lod_tgt_descs *ltd,
+ unsigned idx)
+{
+ LASSERT(LTD_TGT(ltd, idx));
+ if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
+ LTD_TGT(ltd, idx)->ltd_reap = 1;
+ ltd->ltd_death_row++;
+ }
+}
+
+int lod_fini_tgt(struct lod_device *lod, struct lod_tgt_descs *ltd)
+{
+ int idx;
+
+ if (ltd->ltd_tgts_size <= 0)
+ return 0;
+ lod_getref(ltd);
+ mutex_lock(<d->ltd_mutex);
+ cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
+ __lod_del_device(ltd, idx);
+ mutex_unlock(<d->ltd_mutex);
+ lod_putref(lod, ltd);
+ CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
+ for (idx = 0; idx < TGT_PTRS; idx++) {
+ if (ltd->ltd_tgt_idx[idx])
+ OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
+ }
+ ltd->ltd_tgts_size = 0;
+ return 0;
+}
+
+/*
+ * Add support for administratively disabled OST (through the MGS).
+ * Schedule a target for deletion. Disconnection and real removal from the
+ * table takes place in lod_putref() once the last table user release its
+ * reference.
+ *
+ * \param env - is the environment passed by the caller
+ * \param lod - is the lod device currently connected to the OSP about to be
+ * removed
+ * \param osp - is the name of OSP device about to be removed
+ * \param idx - is the OSP index
+ * \param gen - is the generation number, not used currently
+ */
+int lod_del_device(const struct lu_env *env, struct lod_device *lod,
+ struct lod_tgt_descs *ltd, char *osp, unsigned idx,
+ unsigned gen)
+{
+ struct obd_device *obd;
+ int rc = 0;
+ struct obd_uuid uuid;
+ ENTRY;
+
+ CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
+
+ obd_str2uuid(&uuid, osp);
+
+ obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
+ &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
+ if (obd == NULL) {
+ CERROR("can't find %s device\n", osp);
+ RETURN(-EINVAL);
+ }
+
+ if (gen <= 0) {
+ CERROR("%s: request to remove OBD %s with invalid generation %d"
+ "\n", obd->obd_name, osp, gen);
+ RETURN(-EINVAL);
+ }
+
+ obd_str2uuid(&uuid, osp);
+
+ lod_getref(ltd);
+ mutex_lock(<d->ltd_mutex);
+ /* check that the index is allocated in the bitmap */
+ if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
+ !LTD_TGT(ltd, idx)) {
+ CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
+ GOTO(out, rc = -EINVAL);
+ }
+
+ /* check that the UUID matches */
+ if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
+ CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
+ obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
+ idx, osp);
+ GOTO(out, rc = -EINVAL);
+ }
+
+ __lod_del_device(ltd, idx);
+ EXIT;
+out:
+ mutex_unlock(<d->ltd_mutex);
+ lod_putref(lod, ltd);
+ return(rc);
+}
+
+int lod_ea_store_resize(struct lod_thread_info *info, int size)
+{
+ int round = size_roundup_power2(size);
+
+ LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3));
+ if (info->lti_ea_store) {
+ LASSERT(info->lti_ea_store_size);
+ LASSERT(info->lti_ea_store_size < round);
+ CDEBUG(D_INFO, "EA store size %d is not enough, need %d\n",
+ info->lti_ea_store_size, round);
+ OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
+ info->lti_ea_store = NULL;
+ info->lti_ea_store_size = 0;
+ }
+
+ OBD_ALLOC_LARGE(info->lti_ea_store, round);
+ if (info->lti_ea_store == NULL)
+ RETURN(-ENOMEM);
+ info->lti_ea_store_size = round;
+ RETURN(0);
+}
+
+/*
+ * generate and write LOV EA for given striped object
+ */
+int lod_generate_and_set_lovea(const struct lu_env *env,
+ struct lod_object *lo, struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct dt_object *next = dt_object_child(&lo->ldo_obj);
+ const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
+ struct lov_mds_md_v1 *lmm;
+ struct lov_ost_data_v1 *objs;
+ __u32 magic;
+ int i, rc, lmm_size;
+ int cplen = 0;
+ ENTRY;
+
+ LASSERT(lo);
+ LASSERT(lo->ldo_stripenr > 0);
+
+ magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
+ lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
+ if (info->lti_ea_store_size < lmm_size) {
+ rc = lod_ea_store_resize(info, lmm_size);
+ if (rc)
+ RETURN(rc);
+ }
+
+ lmm = info->lti_ea_store;
+
+ lmm->lmm_magic = cpu_to_le32(magic);
+ lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
+ lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
+ lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
+ lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
+ lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
+ lmm->lmm_layout_gen = 0;
+ if (magic == LOV_MAGIC_V1) {
+ objs = &lmm->lmm_objects[0];
+ } else {
+ struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
+ cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
+ sizeof(v3->lmm_pool_name));
+ if (cplen >= sizeof(v3->lmm_pool_name))
+ RETURN(-E2BIG);
+ objs = &v3->lmm_objects[0];
+ }
+
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ const struct lu_fid *fid;
+ struct lod_device *lod;
+ __u32 index;
+
+ lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
+ LASSERT(lo->ldo_stripe[i]);
+ fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
+
+ rc = fid_ostid_pack(fid, &info->lti_ostid);
+ LASSERT(rc == 0);
+
+ objs[i].l_object_id = cpu_to_le64(info->lti_ostid.oi_id);
+ objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
+ objs[i].l_ost_gen = cpu_to_le32(0);
+ rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
+ if (rc < 0) {
+ CERROR("%s: Can not locate "DFID": rc = %d\n",
+ lod2obd(lod)->obd_name, PFID(fid), rc);
+ RETURN(rc);
+ }
+ objs[i].l_ost_idx = cpu_to_le32(index);
+ }
+
+ info->lti_buf.lb_buf = lmm;
+ info->lti_buf.lb_len = lmm_size;
+ rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
+ th, BYPASS_CAPA);
+
+ RETURN(rc);
+}
+
+int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct dt_object *next = dt_object_child(&lo->ldo_obj);
+ int rc;
+ ENTRY;
+
+ LASSERT(info);
+
+ if (unlikely(info->lti_ea_store_size == 0)) {
+ /* just to enter in allocation block below */
+ rc = -ERANGE;
+ } else {
+repeat:
+ info->lti_buf.lb_buf = info->lti_ea_store;
+ info->lti_buf.lb_len = info->lti_ea_store_size;
+ rc = dt_xattr_get(env, next, &info->lti_buf, XATTR_NAME_LOV,
+ BYPASS_CAPA);
+ }
+ /* if object is not striped or inaccessible */
+ if (rc == -ENODATA)
+ RETURN(0);
+
+ if (rc == -ERANGE) {
+ /* EA doesn't fit, reallocate new buffer */
+ rc = dt_xattr_get(env, next, &LU_BUF_NULL, XATTR_NAME_LOV,
+ BYPASS_CAPA);
+ if (rc == -ENODATA)
+ RETURN(0);
+ else if (rc < 0)
+ RETURN(rc);
+
+ LASSERT(rc > 0);
+ rc = lod_ea_store_resize(info, rc);
+ if (rc)
+ RETURN(rc);
+ goto repeat;
+ }
+
+ RETURN(rc);
+}
+
+int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct dt_object *next = dt_object_child(dt);
+ struct lov_user_md_v3 *v3;
+ int rc;
+ int cplen = 0;
+ ENTRY;
+
+ LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
+
+ /*
+ * store striping defaults into new directory
+ * used to implement defaults inheritance
+ */
+
+ /* probably nothing to inherite */
+ if (lo->ldo_striping_cached == 0)
+ RETURN(0);
+
+ if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
+ lo->ldo_def_stripe_offset))
+ RETURN(0);
+
+ /* XXX: use thread info */
+ OBD_ALLOC_PTR(v3);
+ if (v3 == NULL)
+ RETURN(-ENOMEM);
+
+ v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
+ v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
+ v3->lmm_object_id = 0;
+ v3->lmm_object_seq = 0;
+ v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
+ v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
+ v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
+ if (lo->ldo_pool) {
+ cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
+ sizeof(v3->lmm_pool_name));
+ if (cplen >= sizeof(v3->lmm_pool_name)) {
+ OBD_FREE_PTR(v3);
+ RETURN(-E2BIG);