+ ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
+
+ rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
+ (struct dt_rec *)ent,
+ it->lit_attr);
+ if (rc != 0)
+ RETURN(rc);
+
+ /* skip . and .. for slave stripe */
+ if ((strncmp(ent->lde_name, ".",
+ le16_to_cpu(ent->lde_namelen)) == 0 &&
+ le16_to_cpu(ent->lde_namelen) == 1) ||
+ (strncmp(ent->lde_name, "..",
+ le16_to_cpu(ent->lde_namelen)) == 0 &&
+ le16_to_cpu(ent->lde_namelen) == 2))
+ goto again;
+
+ RETURN(rc);
+ }
+
+ /* go to next stripe */
+ if (it->lit_stripe_index + 1 >= lo->ldo_stripenr)
+ RETURN(1);
+
+ it->lit_stripe_index++;
+
+ next->do_index_ops->dio_it.put(env, it->lit_it);
+ next->do_index_ops->dio_it.fini(env, it->lit_it);
+
+ rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
+ if (rc != 0)
+ RETURN(rc);
+
+ next = lo->ldo_stripe[it->lit_stripe_index];
+ LASSERT(next != NULL);
+ LASSERT(next->do_index_ops != NULL);
+
+ it_next = next->do_index_ops->dio_it.init(env, next, it->lit_attr,
+ BYPASS_CAPA);
+ if (!IS_ERR(it_next)) {
+ it->lit_it = it_next;
+ goto again;
+ } else {
+ rc = PTR_ERR(it_next);
+ }
+
+ RETURN(rc);
+}
+
+/**
+ * Implementation of dt_index_operations:: dio_it.key
+ *
+ * This function is to get the key of the iterator at current position.
+ *
+ * \param[in] env execution environment.
+ * \param[in] di the iterator for striped directory.
+ *
+ * \retval key(dt_key) if successfully get the key.
+ * \retval negative error if can not get the key.
+ */
+static struct dt_key *lod_striped_it_key(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ const struct lod_it *it = (const struct lod_it *)di;
+ struct lod_object *lo = lod_dt_obj(it->lit_obj);
+ struct dt_object *next;
+
+ LOD_CHECK_STRIPED_IT(env, it, lo);
+
+ next = lo->ldo_stripe[it->lit_stripe_index];
+ LASSERT(next != NULL);
+ LASSERT(next->do_index_ops != NULL);
+
+ return next->do_index_ops->dio_it.key(env, it->lit_it);
+}
+
+/**
+ * Implementation of dt_index_operations:: dio_it.key_size
+ *
+ * This function is to get the key_size of current key.
+ *
+ * \param[in] env execution environment.
+ * \param[in] di the iterator for striped directory.
+ *
+ * \retval key_size if successfully get the key_size.
+ * \retval negative error if can not get the key_size.
+ */
+static int lod_striped_it_key_size(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ struct lod_it *it = (struct lod_it *)di;
+ struct lod_object *lo = lod_dt_obj(it->lit_obj);
+ struct dt_object *next;
+
+ LOD_CHECK_STRIPED_IT(env, it, lo);
+
+ next = lo->ldo_stripe[it->lit_stripe_index];
+ LASSERT(next != NULL);
+ LASSERT(next->do_index_ops != NULL);
+
+ return next->do_index_ops->dio_it.key_size(env, it->lit_it);
+}
+
+/**
+ * Implementation of dt_index_operations:: dio_it.rec
+ *
+ * This function is to get the record at current position.
+ *
+ * \param[in] env execution environment.
+ * \param[in] di the iterator for striped directory.
+ * \param[in] attr the attribute of iterator, mostly used to indicate
+ * the entry attribute in the object to be iterated.
+ * \param[out] rec hold the return record.
+ *
+ * \retval 0 if successfully get the entry.
+ * \retval negative error if can not get entry.
+ */
+static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
+ struct dt_rec *rec, __u32 attr)
+{
+ const struct lod_it *it = (const struct lod_it *)di;
+ struct lod_object *lo = lod_dt_obj(it->lit_obj);
+ struct dt_object *next;
+
+ LOD_CHECK_STRIPED_IT(env, it, lo);
+
+ next = lo->ldo_stripe[it->lit_stripe_index];
+ LASSERT(next != NULL);
+ LASSERT(next->do_index_ops != NULL);
+
+ return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
+}
+
+/**
+ * Implementation of dt_index_operations:: dio_it.rec_size
+ *
+ * This function is to get the record_size at current record.
+ *
+ * \param[in] env execution environment.
+ * \param[in] di the iterator for striped directory.
+ * \param[in] attr the attribute of iterator, mostly used to indicate
+ * the entry attribute in the object to be iterated.
+ *
+ * \retval rec_size if successfully get the entry size.
+ * \retval negative error if can not get entry size.
+ */
+static int lod_striped_it_rec_size(const struct lu_env *env,
+ const struct dt_it *di, __u32 attr)
+{
+ struct lod_it *it = (struct lod_it *)di;
+ struct lod_object *lo = lod_dt_obj(it->lit_obj);
+ struct dt_object *next;
+
+ LOD_CHECK_STRIPED_IT(env, it, lo);
+
+ next = lo->ldo_stripe[it->lit_stripe_index];
+ LASSERT(next != NULL);
+ LASSERT(next->do_index_ops != NULL);
+
+ return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
+}
+
+/**
+ * Implementation of dt_index_operations:: dio_it.store
+ *
+ * This function will a cookie for current position of the iterator head,
+ * so that user can use this cookie to load/start the iterator next time.
+ *
+ * \param[in] env execution environment.
+ * \param[in] di the iterator for striped directory.
+ *
+ * \retval the cookie.
+ */
+static __u64 lod_striped_it_store(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ const struct lod_it *it = (const struct lod_it *)di;
+ struct lod_object *lo = lod_dt_obj(it->lit_obj);
+ struct dt_object *next;
+
+ LOD_CHECK_STRIPED_IT(env, it, lo);
+
+ next = lo->ldo_stripe[it->lit_stripe_index];
+ LASSERT(next != NULL);
+ LASSERT(next->do_index_ops != NULL);
+
+ return next->do_index_ops->dio_it.store(env, it->lit_it);
+}
+
+/**
+ * Implementation of dt_index_operations:: dio_it.load
+ *
+ * This function will position the iterator with the given hash(usually
+ * get from store),
+ *
+ * \param[in] env execution environment.
+ * \param[in] di the iterator for striped directory.
+ * \param[in] hash the given hash.
+ *
+ * \retval >0 if successfuly load the iterator to the given position.
+ * \retval <0 if load is failed.
+ */
+static int lod_striped_it_load(const struct lu_env *env,
+ const struct dt_it *di, __u64 hash)
+{
+ const struct lod_it *it = (const struct lod_it *)di;
+ struct lod_object *lo = lod_dt_obj(it->lit_obj);
+ struct dt_object *next;
+
+ LOD_CHECK_STRIPED_IT(env, it, lo);
+
+ next = lo->ldo_stripe[it->lit_stripe_index];
+ LASSERT(next != NULL);
+ LASSERT(next->do_index_ops != NULL);
+
+ return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
+}
+
+static struct dt_index_operations lod_striped_index_ops = {
+ .dio_lookup = lod_index_lookup,
+ .dio_declare_insert = lod_declare_index_insert,
+ .dio_insert = lod_index_insert,
+ .dio_declare_delete = lod_declare_index_delete,
+ .dio_delete = lod_index_delete,
+ .dio_it = {
+ .init = lod_striped_it_init,
+ .fini = lod_striped_it_fini,
+ .get = lod_striped_it_get,
+ .put = lod_striped_it_put,
+ .next = lod_striped_it_next,
+ .key = lod_striped_it_key,
+ .key_size = lod_striped_it_key_size,
+ .rec = lod_striped_it_rec,
+ .rec_size = lod_striped_it_rec_size,
+ .store = lod_striped_it_store,
+ .load = lod_striped_it_load,
+ }
+};
+
+/**
+ * Append the FID for each shard of the striped directory after the
+ * given LMV EA header.
+ *
+ * To simplify striped directory and the consistency verification,
+ * we only store the LMV EA header on disk, for both master object
+ * and slave objects. When someone wants to know the whole LMV EA,
+ * such as client readdir(), we can build the entrie LMV EA on the
+ * MDT side (in RAM) via iterating the sub-directory entries that
+ * are contained in the master object of the stripe directory.
+ *
+ * For the master object of the striped directroy, the valid name
+ * for each shard is composed of the ${shard_FID}:${shard_idx}.
+ *
+ * There may be holes in the LMV EA if some shards' name entries
+ * are corrupted or lost.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] lo pointer to the master object of the striped directory
+ * \param[in] buf pointer to the lu_buf which will hold the LMV EA
+ * \param[in] resize whether re-allocate the buffer if it is not big enough
+ *
+ * \retval positive size of the LMV EA
+ * \retval 0 for nothing to be loaded
+ * \retval negative error number on failure
+ */
+int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
+ struct lu_buf *buf, bool resize)
+{
+ struct lu_dirent *ent =
+ (struct lu_dirent *)lod_env_info(env)->lti_key;
+ struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
+ struct dt_object *obj = dt_object_child(&lo->ldo_obj);
+ struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
+ struct dt_it *it;
+ const struct dt_it_ops *iops;
+ __u32 stripes;
+ __u32 magic = le32_to_cpu(lmv1->lmv_magic);
+ size_t lmv1_size;
+ int rc;
+ ENTRY;
+
+ /* If it is not a striped directory, then load nothing. */
+ if (magic != LMV_MAGIC_V1)
+ RETURN(0);
+
+ /* If it is in migration (or failure), then load nothing. */
+ if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
+ RETURN(0);
+
+ stripes = le32_to_cpu(lmv1->lmv_stripe_count);
+ if (stripes < 1)
+ RETURN(0);
+
+ rc = lmv_mds_md_size(stripes, magic);
+ if (rc < 0)
+ RETURN(rc);
+ lmv1_size = rc;
+ if (buf->lb_len < lmv1_size) {
+ struct lu_buf tbuf;
+
+ if (!resize)
+ RETURN(-ERANGE);
+
+ tbuf = *buf;
+ buf->lb_buf = NULL;
+ buf->lb_len = 0;
+ lu_buf_alloc(buf, lmv1_size);
+ lmv1 = buf->lb_buf;
+ if (lmv1 == NULL)
+ RETURN(-ENOMEM);
+
+ memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
+ }
+
+ if (unlikely(!dt_try_as_dir(env, obj)))
+ RETURN(-ENOTDIR);
+
+ memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
+ iops = &obj->do_index_ops->dio_it;
+ it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
+ if (IS_ERR(it))
+ RETURN(PTR_ERR(it));
+
+ rc = iops->load(env, it, 0);
+ if (rc == 0)
+ rc = iops->next(env, it);
+ else if (rc > 0)
+ rc = 0;
+
+ while (rc == 0) {
+ char name[FID_LEN + 2] = "";
+ struct lu_fid fid;
+ __u32 index;
+ int len;
+
+ rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
+ if (rc != 0)
+ break;
+
+ rc = -EIO;
+
+ fid_le_to_cpu(&fid, &ent->lde_fid);
+ ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
+ if (ent->lde_name[0] == '.') {
+ if (ent->lde_namelen == 1)
+ goto next;
+
+ if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
+ goto next;
+ }
+
+ len = snprintf(name, FID_LEN + 1, DFID":", PFID(&ent->lde_fid));
+ /* The ent->lde_name is composed of ${FID}:${index} */
+ if (ent->lde_namelen < len + 1 ||
+ memcmp(ent->lde_name, name, len) != 0) {
+ CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
+ "%s: invalid shard name %.*s with the FID "DFID
+ " for the striped directory "DFID", %s\n",
+ lod2obd(lod)->obd_name, ent->lde_namelen,
+ ent->lde_name, PFID(&fid),
+ PFID(lu_object_fid(&obj->do_lu)),
+ lod->lod_lmv_failout ? "failout" : "skip");
+
+ if (lod->lod_lmv_failout)
+ break;
+
+ goto next;
+ }
+
+ index = 0;
+ do {
+ if (ent->lde_name[len] < '0' ||
+ ent->lde_name[len] > '9') {
+ CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
+ "%s: invalid shard name %.*s with the "
+ "FID "DFID" for the striped directory "
+ DFID", %s\n",
+ lod2obd(lod)->obd_name, ent->lde_namelen,
+ ent->lde_name, PFID(&fid),
+ PFID(lu_object_fid(&obj->do_lu)),
+ lod->lod_lmv_failout ?
+ "failout" : "skip");
+
+ if (lod->lod_lmv_failout)
+ break;
+
+ goto next;
+ }
+
+ index = index * 10 + ent->lde_name[len++] - '0';
+ } while (len < ent->lde_namelen);
+
+ if (len == ent->lde_namelen) {
+ /* Out of LMV EA range. */
+ if (index >= stripes) {
+ CERROR("%s: the shard %.*s for the striped "
+ "directory "DFID" is out of the known "
+ "LMV EA range [0 - %u], failout\n",
+ lod2obd(lod)->obd_name, ent->lde_namelen,
+ ent->lde_name,
+ PFID(lu_object_fid(&obj->do_lu)),
+ stripes - 1);
+
+ break;
+ }
+
+ /* The slot has been occupied. */
+ if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
+ struct lu_fid fid0;
+
+ fid_le_to_cpu(&fid0,
+ &lmv1->lmv_stripe_fids[index]);
+ CERROR("%s: both the shard "DFID" and "DFID
+ " for the striped directory "DFID
+ " claim the same LMV EA slot at the "
+ "index %d, failout\n",
+ lod2obd(lod)->obd_name,
+ PFID(&fid0), PFID(&fid),
+ PFID(lu_object_fid(&obj->do_lu)), index);
+
+ break;
+ }
+
+ /* stored as LE mode */
+ lmv1->lmv_stripe_fids[index] = ent->lde_fid;
+
+next:
+ rc = iops->next(env, it);
+ }
+ }
+
+ iops->put(env, it);
+ iops->fini(env, it);
+
+ RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
+}
+
+/**
+ * Implementation of dt_object_operations:: do_index_try
+ *
+ * This function will try to initialize the index api pointer for the
+ * given object, usually it the entry point of the index api. i.e.
+ * the index object should be initialized in index_try, then start
+ * using index api. For striped directory, it will try to initialize
+ * all of its sub_stripes.
+ *
+ * \param[in] env execution environment.
+ * \param[in] dt the index object to be initialized.
+ * \param[in] feat the features of this object, for example fixed or
+ * variable key size etc.
+ *
+ * \retval >0 if the initialization is successful.
+ * \retval <0 if the initialization is failed.
+ */
+static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
+ const struct dt_index_features *feat)
+{
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct dt_object *next = dt_object_child(dt);
+ int rc;
+ ENTRY;
+
+ LASSERT(next->do_ops);
+ LASSERT(next->do_ops->do_index_try);
+
+ rc = lod_load_striping_locked(env, lo);
+ if (rc != 0)
+ RETURN(rc);
+
+ rc = next->do_ops->do_index_try(env, next, feat);
+ if (rc != 0)
+ RETURN(rc);
+
+ if (lo->ldo_stripenr > 0) {
+ int i;
+
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ if (dt_object_exists(lo->ldo_stripe[i]) == 0)
+ continue;
+ rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
+ lo->ldo_stripe[i], feat);
+ if (rc != 0)
+ RETURN(rc);
+ }
+ dt->do_index_ops = &lod_striped_index_ops;
+ } else {
+ dt->do_index_ops = &lod_index_ops;
+ }
+
+ RETURN(rc);
+}
+
+static void lod_object_read_lock(const struct lu_env *env,
+ struct dt_object *dt, unsigned role)
+{
+ dt_read_lock(env, dt_object_child(dt), role);
+}
+
+static void lod_object_write_lock(const struct lu_env *env,
+ struct dt_object *dt, unsigned role)
+{
+ dt_write_lock(env, dt_object_child(dt), role);
+}
+
+static void lod_object_read_unlock(const struct lu_env *env,
+ struct dt_object *dt)
+{
+ dt_read_unlock(env, dt_object_child(dt));
+}
+
+static void lod_object_write_unlock(const struct lu_env *env,
+ struct dt_object *dt)
+{
+ dt_write_unlock(env, dt_object_child(dt));
+}
+
+static int lod_object_write_locked(const struct lu_env *env,
+ struct dt_object *dt)
+{
+ return dt_write_locked(env, dt_object_child(dt));
+}
+
+static int lod_attr_get(const struct lu_env *env,
+ struct dt_object *dt,
+ struct lu_attr *attr,
+ struct lustre_capa *capa)
+{
+ /* Note: for striped directory, client will merge attributes
+ * from all of the sub-stripes see lmv_merge_attr(), and there
+ * no MDD logic depend on directory nlink/size/time, so we can
+ * always use master inode nlink and size for now. */
+ return dt_attr_get(env, dt_object_child(dt), attr, capa);
+}
+
+/**
+ * Mark all of sub-stripes dead of the striped directory.
+ **/
+static int lod_mark_dead_object(const struct lu_env *env,
+ struct dt_object *dt,
+ struct thandle *handle,
+ bool declare)
+{
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct lmv_mds_md_v1 *lmv;
+ __u32 dead_hash_type;
+ int rc;
+ int i;
+
+ ENTRY;
+
+ if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
+ RETURN(0);
+
+ rc = lod_load_striping_locked(env, lo);
+ if (rc != 0)
+ RETURN(rc);
+
+ if (lo->ldo_stripenr == 0)
+ RETURN(0);
+
+ rc = lod_get_lmv_ea(env, lo);
+ if (rc <= 0)
+ RETURN(rc);
+
+ lmv = lod_env_info(env)->lti_ea_store;
+ lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
+ dead_hash_type = le32_to_cpu(lmv->lmv_hash_type) | LMV_HASH_FLAG_DEAD;
+ lmv->lmv_hash_type = cpu_to_le32(dead_hash_type);
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ struct lu_buf buf;
+
+ lmv->lmv_master_mdt_index = i;
+ buf.lb_buf = lmv;
+ buf.lb_len = sizeof(*lmv);
+ if (declare) {
+ rc = dt_declare_xattr_set(env, lo->ldo_stripe[i], &buf,
+ XATTR_NAME_LMV,
+ LU_XATTR_REPLACE, handle);
+ } else {
+ rc = dt_xattr_set(env, lo->ldo_stripe[i], &buf,
+ XATTR_NAME_LMV, LU_XATTR_REPLACE,
+ handle, BYPASS_CAPA);
+ }
+ if (rc != 0)
+ break;
+ }
+
+ RETURN(rc);
+}
+
+static int lod_declare_attr_set(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_attr *attr,
+ struct thandle *handle)
+{
+ struct dt_object *next = dt_object_child(dt);
+ struct lod_object *lo = lod_dt_obj(dt);
+ int rc, i;
+ ENTRY;
+
+ /* Set dead object on all other stripes */
+ if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
+ attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
+ rc = lod_mark_dead_object(env, dt, handle, true);
+ RETURN(rc);
+ }
+
+ /*
+ * declare setattr on the local object
+ */
+ rc = dt_declare_attr_set(env, next, attr, handle);
+ if (rc)
+ RETURN(rc);
+
+ /* osp_declare_attr_set() ignores all attributes other than
+ * UID, GID, and size, and osp_attr_set() ignores all but UID
+ * and GID. Declaration of size attr setting happens through
+ * lod_declare_init_size(), and not through this function.
+ * Therefore we need not load striping unless ownership is
+ * changing. This should save memory and (we hope) speed up
+ * rename(). */
+ if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
+ if (!(attr->la_valid & (LA_UID | LA_GID)))
+ RETURN(rc);
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
+ RETURN(0);
+ } else {
+ if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
+ LA_ATIME | LA_MTIME | LA_CTIME)))
+ RETURN(rc);
+ }
+ /*
+ * load striping information, notice we don't do this when object
+ * is being initialized as we don't need this information till
+ * few specific cases like destroy, chown
+ */
+ rc = lod_load_striping(env, lo);
+ if (rc)
+ RETURN(rc);
+
+ if (lo->ldo_stripenr == 0)
+ RETURN(0);
+
+ /*
+ * if object is striped declare changes on the stripes
+ */
+ LASSERT(lo->ldo_stripe);
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ if (likely(lo->ldo_stripe[i] != NULL)) {
+ rc = dt_declare_attr_set(env, lo->ldo_stripe[i], attr,
+ handle);
+ if (rc != 0) {
+ CERROR("failed declaration: %d\n", rc);
+ break;
+ }
+ }
+ }
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
+ dt_object_exists(next) != 0 &&
+ dt_object_remote(next) == 0)
+ dt_declare_xattr_del(env, next, XATTR_NAME_LOV, handle);
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
+ dt_object_exists(next) &&
+ dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lu_buf *buf = &info->lti_buf;
+
+ buf->lb_buf = info->lti_ea_store;
+ buf->lb_len = info->lti_ea_store_size;
+ dt_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
+ LU_XATTR_REPLACE, handle);
+ }
+
+ RETURN(rc);
+}
+
+static int lod_attr_set(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_attr *attr,
+ struct thandle *handle,
+ struct lustre_capa *capa)
+{
+ struct dt_object *next = dt_object_child(dt);
+ struct lod_object *lo = lod_dt_obj(dt);
+ int rc, i;
+ ENTRY;
+
+ /* Set dead object on all other stripes */
+ if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
+ attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
+ rc = lod_mark_dead_object(env, dt, handle, false);
+ RETURN(rc);
+ }
+
+ /*
+ * apply changes to the local object
+ */
+ rc = dt_attr_set(env, next, attr, handle, capa);
+ if (rc)
+ RETURN(rc);
+
+ if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
+ if (!(attr->la_valid & (LA_UID | LA_GID)))
+ RETURN(rc);
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
+ RETURN(0);
+ } else {
+ if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
+ LA_ATIME | LA_MTIME | LA_CTIME)))
+ RETURN(rc);
+ }
+
+ if (lo->ldo_stripenr == 0)
+ RETURN(0);
+
+ /*
+ * if object is striped, apply changes to all the stripes
+ */
+ LASSERT(lo->ldo_stripe);
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ if (unlikely(lo->ldo_stripe[i] == NULL))
+ continue;
+ if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
+ (dt_object_exists(lo->ldo_stripe[i]) == 0))
+ continue;
+
+ rc = dt_attr_set(env, lo->ldo_stripe[i], attr, handle, capa);
+ if (rc != 0) {
+ CERROR("failed declaration: %d\n", rc);
+ break;
+ }
+ }
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
+ dt_object_exists(next) != 0 &&
+ dt_object_remote(next) == 0)
+ dt_xattr_del(env, next, XATTR_NAME_LOV, handle, BYPASS_CAPA);
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
+ dt_object_exists(next) &&
+ dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lu_buf *buf = &info->lti_buf;
+ struct ost_id *oi = &info->lti_ostid;
+ struct lu_fid *fid = &info->lti_fid;
+ struct lov_mds_md_v1 *lmm;
+ struct lov_ost_data_v1 *objs;
+ __u32 magic;
+ int rc1;
+
+ rc1 = lod_get_lov_ea(env, lo);
+ if (rc1 <= 0)
+ RETURN(rc);
+
+ buf->lb_buf = info->lti_ea_store;
+ buf->lb_len = info->lti_ea_store_size;
+ lmm = info->lti_ea_store;
+ magic = le32_to_cpu(lmm->lmm_magic);
+ if (magic == LOV_MAGIC_V1)
+ objs = &(lmm->lmm_objects[0]);
+ else
+ objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
+ ostid_le_to_cpu(&objs->l_ost_oi, oi);
+ ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
+ fid->f_oid--;
+ fid_to_ostid(fid, oi);
+ ostid_cpu_to_le(oi, &objs->l_ost_oi);
+ dt_xattr_set(env, next, buf, XATTR_NAME_LOV,
+ LU_XATTR_REPLACE, handle, BYPASS_CAPA);
+ }
+
+ RETURN(rc);
+}
+
+static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
+ struct lu_buf *buf, const char *name,
+ struct lustre_capa *capa)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
+ int rc, is_root;
+ ENTRY;
+
+ rc = dt_xattr_get(env, dt_object_child(dt), buf, name, capa);
+ if (strcmp(name, XATTR_NAME_LMV) == 0) {
+ struct lmv_mds_md_v1 *lmv1;
+ int rc1 = 0;
+
+ if (rc > (typeof(rc))sizeof(*lmv1))
+ RETURN(rc);
+
+ if (rc < (typeof(rc))sizeof(*lmv1))
+ RETURN(rc = rc > 0 ? -EINVAL : rc);
+
+ if (buf->lb_buf == NULL || buf->lb_len == 0) {
+ CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
+
+ info->lti_buf.lb_buf = info->lti_key;
+ info->lti_buf.lb_len = sizeof(*lmv1);
+ rc = dt_xattr_get(env, dt_object_child(dt),
+ &info->lti_buf, name, capa);
+ if (unlikely(rc != sizeof(*lmv1)))
+ RETURN(rc = rc > 0 ? -EINVAL : rc);
+
+ lmv1 = info->lti_buf.lb_buf;
+ /* The on-disk LMV EA only contains header, but the
+ * returned LMV EA size should contain the space for
+ * the FIDs of all shards of the striped directory. */
+ if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
+ rc = lmv_mds_md_size(
+ le32_to_cpu(lmv1->lmv_stripe_count),
+ LMV_MAGIC_V1);
+ } else {
+ rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
+ buf, false);
+ }
+
+ RETURN(rc = rc1 != 0 ? rc1 : rc);
+ }
+
+ if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
+ RETURN(rc);
+
+ /*
+ * lod returns default striping on the real root of the device
+ * this is like the root stores default striping for the whole
+ * filesystem. historically we've been using a different approach
+ * and store it in the config.
+ */
+ dt_root_get(env, dev->lod_child, &info->lti_fid);
+ is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
+
+ if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
+ struct lov_user_md *lum = buf->lb_buf;
+ struct lov_desc *desc = &dev->lod_desc;
+
+ if (buf->lb_buf == NULL) {
+ rc = sizeof(*lum);
+ } else if (buf->lb_len >= sizeof(*lum)) {
+ lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
+ lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
+ lmm_oi_set_id(&lum->lmm_oi, 0);
+ lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
+ lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
+ lum->lmm_stripe_size = cpu_to_le32(
+ desc->ld_default_stripe_size);
+ lum->lmm_stripe_count = cpu_to_le16(
+ desc->ld_default_stripe_count);
+ lum->lmm_stripe_offset = cpu_to_le16(
+ desc->ld_default_stripe_offset);
+ rc = sizeof(*lum);
+ } else {
+ rc = -ERANGE;
+ }
+ }
+
+ RETURN(rc);
+}
+
+static int lod_verify_md_striping(struct lod_device *lod,
+ const struct lmv_user_md_v1 *lum)
+{
+ int rc = 0;
+ ENTRY;
+
+ if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC))
+ GOTO(out, rc = -EINVAL);
+
+ if (unlikely(le32_to_cpu(lum->lum_stripe_count) == 0))
+ GOTO(out, rc = -EINVAL);
+out:
+ if (rc != 0)
+ CERROR("%s: invalid lmv_user_md: magic = %x, "
+ "stripe_offset = %d, stripe_count = %u: rc = %d\n",
+ lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
+ (int)le32_to_cpu(lum->lum_stripe_offset),
+ le32_to_cpu(lum->lum_stripe_count), rc);
+ return rc;
+}
+
+/**
+ * Master LMVEA will be same as slave LMVEA, except
+ * 1. different magic
+ * 2. lmv_master_mdt_index on slave LMV EA will be stripe_index.
+ */
+static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
+ const struct lmv_mds_md_v1 *master_lmv)
+{
+ *slave_lmv = *master_lmv;
+ slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
+}
+
+int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
+ struct lu_buf *lmv_buf)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct lmv_mds_md_v1 *lmm1;
+ int stripe_count;
+ int type = LU_SEQ_RANGE_ANY;
+ int rc;
+ __u32 mdtidx;
+ ENTRY;
+
+ LASSERT(lo->ldo_dir_striped != 0);
+ LASSERT(lo->ldo_stripenr > 0);
+ stripe_count = lo->ldo_stripenr;
+ /* Only store the LMV EA heahder on the disk. */
+ if (info->lti_ea_store_size < sizeof(*lmm1)) {
+ rc = lod_ea_store_resize(info, sizeof(*lmm1));
+ if (rc != 0)
+ RETURN(rc);
+ } else {
+ memset(info->lti_ea_store, 0, sizeof(*lmm1));
+ }
+
+ lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
+ lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
+ lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
+ lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
+ rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
+ &mdtidx, &type);
+ if (rc != 0)
+ RETURN(rc);
+
+ lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
+ lmv_buf->lb_buf = info->lti_ea_store;
+ lmv_buf->lb_len = sizeof(*lmm1);
+ lo->ldo_dir_striping_cached = 1;
+
+ RETURN(rc);
+}
+
+int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
+ const struct lu_buf *buf)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
+ struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
+ struct dt_object **stripe;
+ union lmv_mds_md *lmm = buf->lb_buf;
+ struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
+ struct lu_fid *fid = &info->lti_fid;
+ unsigned int i;
+ int rc = 0;
+ ENTRY;
+
+ if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
+ RETURN(0);
+
+ if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
+ lo->ldo_dir_slave_stripe = 1;
+ RETURN(0);
+ }
+
+ if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
+ RETURN(-EINVAL);
+
+ if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
+ RETURN(0);
+
+ LASSERT(lo->ldo_stripe == NULL);
+ OBD_ALLOC(stripe, sizeof(stripe[0]) *
+ (le32_to_cpu(lmv1->lmv_stripe_count)));
+ if (stripe == NULL)
+ RETURN(-ENOMEM);
+
+ for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
+ struct dt_device *tgt_dt;
+ struct dt_object *dto;
+ int type = LU_SEQ_RANGE_ANY;
+ __u32 idx;
+
+ fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
+ if (!fid_is_sane(fid))
+ GOTO(out, rc = -ESTALE);
+
+ rc = lod_fld_lookup(env, lod, fid, &idx, &type);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
+ tgt_dt = lod->lod_child;
+ } else {
+ struct lod_tgt_desc *tgt;
+
+ tgt = LTD_TGT(ltd, idx);
+ if (tgt == NULL)
+ GOTO(out, rc = -ESTALE);
+ tgt_dt = tgt->ltd_tgt;
+ }
+
+ dto = dt_locate_at(env, tgt_dt, fid,
+ lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
+ NULL);
+ if (IS_ERR(dto))
+ GOTO(out, rc = PTR_ERR(dto));
+
+ stripe[i] = dto;
+ }
+out:
+ lo->ldo_stripe = stripe;
+ lo->ldo_stripenr = le32_to_cpu(lmv1->lmv_stripe_count);
+ lo->ldo_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
+ if (rc != 0)
+ lod_object_free_striping(env, lo);
+
+ RETURN(rc);
+}
+
+static int lod_prep_md_striped_create(const struct lu_env *env,
+ struct dt_object *dt,
+ struct lu_attr *attr,
+ const struct lmv_user_md_v1 *lum,
+ struct dt_object_format *dof,
+ struct thandle *th)
+{
+ struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
+ struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct lod_thread_info *info = lod_env_info(env);
+ struct dt_object **stripe;
+ struct lu_buf lmv_buf;
+ struct lu_buf slave_lmv_buf;
+ struct lmv_mds_md_v1 *lmm;
+ struct lmv_mds_md_v1 *slave_lmm = NULL;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ __u32 stripe_count;
+ int *idx_array;
+ int rc = 0;
+ __u32 i;
+ __u32 j;
+ ENTRY;
+
+ /* The lum has been verifed in lod_verify_md_striping */
+ LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
+ LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
+
+ stripe_count = le32_to_cpu(lum->lum_stripe_count);
+
+ /* shrink the stripe_count to the avaible MDT count */
+ if (stripe_count > lod->lod_remote_mdt_count + 1)
+ stripe_count = lod->lod_remote_mdt_count + 1;
+
+ OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
+ if (stripe == NULL)
+ RETURN(-ENOMEM);
+
+ OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
+ if (idx_array == NULL)
+ GOTO(out_free, rc = -ENOMEM);
+
+ for (i = 0; i < stripe_count; i++) {
+ struct lod_tgt_desc *tgt = NULL;
+ struct dt_object *dto;
+ struct lu_fid fid = { 0 };
+ int idx;
+ struct lu_object_conf conf = { 0 };
+ struct dt_device *tgt_dt = NULL;
+
+ if (i == 0) {
+ /* Right now, master stripe and master object are
+ * on the same MDT */
+ idx = le32_to_cpu(lum->lum_stripe_offset);
+ rc = obd_fid_alloc(env, lod->lod_child_exp, &fid,
+ NULL);
+ if (rc < 0)
+ GOTO(out_put, rc);
+ tgt_dt = lod->lod_child;
+ goto next;
+ }
+
+ idx = (idx_array[i - 1] + 1) % (lod->lod_remote_mdt_count + 1);
+
+ for (j = 0; j < lod->lod_remote_mdt_count;
+ j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
+ bool already_allocated = false;
+ __u32 k;
+
+ CDEBUG(D_INFO, "try idx %d, mdt cnt %u,"
+ " allocated %u, last allocated %d\n", idx,
+ lod->lod_remote_mdt_count, i, idx_array[i - 1]);
+
+ /* Find next available target */
+ if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx))
+ continue;
+
+ /* check whether the idx already exists
+ * in current allocated array */
+ for (k = 0; k < i; k++) {
+ if (idx_array[k] == idx) {
+ already_allocated = true;
+ break;
+ }
+ }
+
+ if (already_allocated)
+ continue;
+
+ /* check the status of the OSP */
+ tgt = LTD_TGT(ltd, idx);
+ if (tgt == NULL)
+ continue;
+
+ tgt_dt = tgt->ltd_tgt;
+ rc = dt_statfs(env, tgt_dt, NULL);
+ if (rc) {
+ /* this OSP doesn't feel well */
+ rc = 0;
+ continue;
+ }
+
+ rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
+ if (rc < 0) {
+ rc = 0;
+ continue;
+ }
+
+ break;
+ }
+
+ /* Can not allocate more stripes */
+ if (j == lod->lod_remote_mdt_count) {
+ CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
+ lod2obd(lod)->obd_name, stripe_count, i - 1);
+ break;
+ }
+
+ CDEBUG(D_INFO, "idx %d, mdt cnt %u,"
+ " allocated %u, last allocated %d\n", idx,
+ lod->lod_remote_mdt_count, i, idx_array[i - 1]);
+
+next:
+ /* tgt_dt and fid must be ready after search avaible OSP
+ * in the above loop */
+ LASSERT(tgt_dt != NULL);
+ LASSERT(fid_is_sane(&fid));
+ conf.loc_flags = LOC_F_NEW;
+ dto = dt_locate_at(env, tgt_dt, &fid,
+ dt->do_lu.lo_dev->ld_site->ls_top_dev,
+ &conf);
+ if (IS_ERR(dto))
+ GOTO(out_put, rc = PTR_ERR(dto));
+ stripe[i] = dto;
+ idx_array[i] = idx;
+ }
+
+ lo->ldo_dir_striped = 1;
+ lo->ldo_stripe = stripe;
+ lo->ldo_stripenr = i;
+ lo->ldo_stripes_allocated = stripe_count;
+
+ if (lo->ldo_stripenr == 0)
+ GOTO(out_put, rc = -ENOSPC);
+
+ rc = lod_prep_lmv_md(env, dt, &lmv_buf);
+ if (rc != 0)
+ GOTO(out_put, rc);
+ lmm = lmv_buf.lb_buf;
+
+ OBD_ALLOC_PTR(slave_lmm);
+ if (slave_lmm == NULL)
+ GOTO(out_put, rc = -ENOMEM);
+
+ lod_prep_slave_lmv_md(slave_lmm, lmm);
+ slave_lmv_buf.lb_buf = slave_lmm;
+ slave_lmv_buf.lb_len = sizeof(*slave_lmm);
+
+ if (!dt_try_as_dir(env, dt_object_child(dt)))
+ GOTO(out_put, rc = -EINVAL);
+
+ rec->rec_type = S_IFDIR;
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ struct dt_object *dto = stripe[i];
+ char *stripe_name = info->lti_key;
+ struct lu_name *sname;
+ struct linkea_data ldata = { 0 };
+ struct lu_buf linkea_buf;
+
+ rc = dt_declare_create(env, dto, attr, NULL, dof, th);
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+ if (!dt_try_as_dir(env, dto))
+ GOTO(out_put, rc = -EINVAL);
+
+ rec->rec_fid = lu_object_fid(&dto->do_lu);
+ rc = dt_declare_insert(env, dto, (const struct dt_rec *)rec,
+ (const struct dt_key *)dot, th);
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+ /* master stripe FID will be put to .. */
+ rec->rec_fid = lu_object_fid(&dt->do_lu);
+ rc = dt_declare_insert(env, dto, (const struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th);
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+ /* probably nothing to inherite */
+ if (lo->ldo_striping_cached &&
+ !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
+ lo->ldo_def_stripenr,
+ lo->ldo_def_stripe_offset)) {
+ struct lov_user_md_v3 *v3;
+
+ /* sigh, lti_ea_store has been used for lmv_buf,
+ * so we have to allocate buffer for default
+ * stripe EA */
+ OBD_ALLOC_PTR(v3);
+ if (v3 == NULL)
+ GOTO(out_put, rc = -ENOMEM);
+
+ memset(v3, 0, sizeof(*v3));
+ v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
+ v3->lmm_stripe_count =
+ cpu_to_le16(lo->ldo_def_stripenr);
+ v3->lmm_stripe_offset =
+ cpu_to_le16(lo->ldo_def_stripe_offset);
+ v3->lmm_stripe_size =
+ cpu_to_le32(lo->ldo_def_stripe_size);
+ if (lo->ldo_pool != NULL)
+ strlcpy(v3->lmm_pool_name, lo->ldo_pool,
+ sizeof(v3->lmm_pool_name));
+
+ info->lti_buf.lb_buf = v3;
+ info->lti_buf.lb_len = sizeof(*v3);
+ rc = dt_declare_xattr_set(env, dto,
+ &info->lti_buf,
+ XATTR_NAME_LOV,
+ 0, th);
+ OBD_FREE_PTR(v3);
+ if (rc != 0)
+ GOTO(out_put, rc);
+ }
+
+ slave_lmm->lmv_master_mdt_index = cpu_to_le32(i);
+ rc = dt_declare_xattr_set(env, dto, &slave_lmv_buf,
+ XATTR_NAME_LMV, 0, th);
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+ snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
+ PFID(lu_object_fid(&dto->do_lu)), i);
+
+ sname = lod_name_get(env, stripe_name, strlen(stripe_name));
+ rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+ rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+ linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
+ linkea_buf.lb_len = ldata.ld_leh->leh_len;
+ rc = dt_declare_xattr_set(env, dto, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+ rec->rec_fid = lu_object_fid(&dto->do_lu);
+ rc = dt_declare_insert(env, dt_object_child(dt),
+ (const struct dt_rec *)rec,
+ (const struct dt_key *)stripe_name, th);
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+ rc = dt_declare_ref_add(env, dt_object_child(dt), th);
+ if (rc != 0)
+ GOTO(out_put, rc);
+ }
+
+ rc = dt_declare_xattr_set(env, dt_object_child(dt), &lmv_buf,
+ XATTR_NAME_LMV, 0, th);
+ if (rc != 0)
+ GOTO(out_put, rc);
+
+out_put:
+ if (rc < 0) {
+ for (i = 0; i < stripe_count; i++)
+ if (stripe[i] != NULL)
+ lu_object_put(env, &stripe[i]->do_lu);
+ OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
+ lo->ldo_stripenr = 0;
+ lo->ldo_stripes_allocated = 0;
+ lo->ldo_stripe = NULL;
+ }
+
+out_free:
+ if (idx_array != NULL)
+ OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
+ if (slave_lmm != NULL)
+ OBD_FREE_PTR(slave_lmm);
+
+ RETURN(rc);
+}
+
+/**
+ * Declare create striped md object.
+ */
+static int lod_declare_xattr_set_lmv(const struct lu_env *env,
+ struct dt_object *dt,
+ struct lu_attr *attr,
+ const struct lu_buf *lum_buf,
+ struct dt_object_format *dof,
+ struct thandle *th)
+{
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
+ struct lmv_user_md_v1 *lum;
+ int rc;
+ ENTRY;
+
+ lum = lum_buf->lb_buf;
+ LASSERT(lum != NULL);
+
+ CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
+ le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
+ (int)le32_to_cpu(lum->lum_stripe_offset));
+
+ if (le32_to_cpu(lum->lum_stripe_count) == 0)
+ GOTO(out, rc = 0);
+
+ rc = lod_verify_md_striping(lod, lum);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ /* prepare dir striped objects */
+ rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
+ if (rc != 0) {
+ /* failed to create striping, let's reset
+ * config so that others don't get confused */
+ lod_object_free_striping(env, lo);
+ GOTO(out, rc);
+ }
+out:
+ RETURN(rc);
+}
+
+static int lod_dir_declare_xattr_set(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_buf *buf,
+ const char *name, int fl,
+ struct thandle *th)
+{
+ struct dt_object *next = dt_object_child(dt);
+ struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
+ struct lod_object *lo = lod_dt_obj(dt);
+ int i;
+ int rc;
+ ENTRY;
+
+ if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
+ struct lmv_user_md_v1 *lum;
+
+ LASSERT(buf != NULL && buf->lb_buf != NULL);
+ lum = buf->lb_buf;
+ rc = lod_verify_md_striping(d, lum);
+ if (rc != 0)
+ RETURN(rc);
+ }
+
+ rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
+ if (rc != 0)
+ RETURN(rc);
+
+ /* set xattr to each stripes, if needed */
+ rc = lod_load_striping(env, lo);
+ if (rc != 0)
+ RETURN(rc);
+
+ /* Note: Do not set LinkEA on sub-stripes, otherwise
+ * it will confuse the fid2path process(see mdt_path_current()).
+ * The linkEA between master and sub-stripes is set in
+ * lod_xattr_set_lmv(). */
+ if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
+ RETURN(0);
+
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ LASSERT(lo->ldo_stripe[i]);
+ rc = dt_declare_xattr_set(env, lo->ldo_stripe[i], buf,
+ name, fl, th);
+ if (rc != 0)
+ break;
+ }
+
+ RETURN(rc);
+}
+
+/*
+ * LOV xattr is a storage for striping, and LOD owns this xattr.
+ * but LOD allows others to control striping to some extent
+ * - to reset strping
+ * - to set new defined striping
+ * - to set new semi-defined striping
+ * - number of stripes is defined
+ * - number of stripes + osts are defined
+ * - ??
+ */
+static int lod_declare_xattr_set(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_buf *buf,
+ const char *name, int fl,
+ struct thandle *th)
+{
+ struct dt_object *next = dt_object_child(dt);
+ struct lu_attr *attr = &lod_env_info(env)->lti_attr;
+ __u32 mode;
+ int rc;
+ ENTRY;
+
+ /*
+ * allow to declare predefined striping on a new (!mode) object
+ * which is supposed to be replay of regular file creation
+ * (when LOV setting is declared)
+ * LU_XATTR_REPLACE is set to indicate a layout swap
+ */
+ mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
+ if ((S_ISREG(mode) || mode == 0) && strcmp(name, XATTR_NAME_LOV) == 0 &&
+ !(fl & LU_XATTR_REPLACE)) {
+ /*
+ * this is a request to manipulate object's striping
+ */
+ if (dt_object_exists(dt)) {
+ rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
+ if (rc)
+ RETURN(rc);
+ } else {
+ memset(attr, 0, sizeof(*attr));
+ attr->la_valid = LA_TYPE | LA_MODE;
+ attr->la_mode = S_IFREG;
+ }
+ rc = lod_declare_striped_object(env, dt, attr, buf, th);
+ } else if (S_ISDIR(mode)) {
+ rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
+ } else {
+ rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
+ }
+
+ RETURN(rc);
+}
+
+static void lod_lov_stripe_cache_clear(struct lod_object *lo)
+{
+ lo->ldo_striping_cached = 0;
+ lo->ldo_def_striping_set = 0;
+ lod_object_set_pool(lo, NULL);
+ lo->ldo_def_stripe_size = 0;
+ lo->ldo_def_stripenr = 0;
+ if (lo->ldo_dir_stripe != NULL)
+ lo->ldo_dir_striping_cached = 0;
+}
+
+static int lod_xattr_set_internal(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_buf *buf,
+ const char *name, int fl, struct thandle *th,
+ struct lustre_capa *capa)
+{
+ struct dt_object *next = dt_object_child(dt);
+ struct lod_object *lo = lod_dt_obj(dt);
+ int rc;
+ int i;
+ ENTRY;
+
+ rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
+ if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
+ RETURN(rc);
+
+ /* Note: Do not set LinkEA on sub-stripes, otherwise
+ * it will confuse the fid2path process(see mdt_path_current()).
+ * The linkEA between master and sub-stripes is set in
+ * lod_xattr_set_lmv(). */
+ if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
+ RETURN(0);
+
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ LASSERT(lo->ldo_stripe[i]);
+ rc = dt_xattr_set(env, lo->ldo_stripe[i], buf, name, fl, th,
+ capa);
+ if (rc != 0)
+ break;
+ }
+
+ RETURN(rc);
+}
+
+static int lod_xattr_del_internal(const struct lu_env *env,
+ struct dt_object *dt,
+ const char *name, struct thandle *th,
+ struct lustre_capa *capa)
+{
+ struct dt_object *next = dt_object_child(dt);
+ struct lod_object *lo = lod_dt_obj(dt);
+ int rc;
+ int i;
+ ENTRY;
+
+ rc = dt_xattr_del(env, next, name, th, capa);
+ if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
+ RETURN(rc);
+
+ if (lo->ldo_stripenr == 0)
+ RETURN(rc);
+
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ LASSERT(lo->ldo_stripe[i]);
+ rc = dt_xattr_del(env, lo->ldo_stripe[i], name, th,
+ capa);
+ if (rc != 0)
+ break;
+ }
+
+ RETURN(rc);
+}
+
+static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_buf *buf,
+ const char *name, int fl,
+ struct thandle *th,
+ struct lustre_capa *capa)
+{
+ struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
+ struct lod_object *l = lod_dt_obj(dt);
+ struct lov_user_md_v1 *lum;
+ struct lov_user_md_v3 *v3 = NULL;
+ int rc;
+ ENTRY;
+
+ /* If it is striped dir, we should clear the stripe cache for
+ * slave stripe as well, but there are no effective way to
+ * notify the LOD on the slave MDT, so we do not cache stripe
+ * information for slave stripe for now. XXX*/
+ lod_lov_stripe_cache_clear(l);
+ LASSERT(buf != NULL && buf->lb_buf != NULL);
+ lum = buf->lb_buf;
+
+ rc = lod_verify_striping(d, buf, false);
+ if (rc)
+ RETURN(rc);
+
+ if (lum->lmm_magic == LOV_USER_MAGIC_V3)
+ v3 = buf->lb_buf;
+
+ /* if { size, offset, count } = { 0, -1, 0 } and no pool
+ * (i.e. all default values specified) then delete default
+ * striping from dir. */
+ CDEBUG(D_OTHER,
+ "set default striping: sz %u # %u offset %d %s %s\n",
+ (unsigned)lum->lmm_stripe_size,
+ (unsigned)lum->lmm_stripe_count,
+ (int)lum->lmm_stripe_offset,
+ v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
+
+ if (LOVEA_DELETE_VALUES((lum->lmm_stripe_size),
+ (lum->lmm_stripe_count),
+ (lum->lmm_stripe_offset)) &&
+ lum->lmm_magic == LOV_USER_MAGIC_V1) {
+ rc = lod_xattr_del_internal(env, dt, name, th, capa);
+ if (rc == -ENODATA)
+ rc = 0;
+ } else {
+ rc = lod_xattr_set_internal(env, dt, buf, name, fl, th, capa);
+ }
+
+ RETURN(rc);
+}
+
+static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_buf *buf,
+ const char *name, int fl,
+ struct thandle *th,
+ struct lustre_capa *capa)
+{
+ struct lod_object *l = lod_dt_obj(dt);
+ struct lmv_user_md_v1 *lum;
+ int rc;
+ ENTRY;
+
+ LASSERT(buf != NULL && buf->lb_buf != NULL);
+ lum = buf->lb_buf;
+
+ CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
+ le32_to_cpu(lum->lum_stripe_count),
+ (int)le32_to_cpu(lum->lum_stripe_offset));
+
+ if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
+ le32_to_cpu(lum->lum_stripe_offset)) &&
+ le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
+ rc = lod_xattr_del_internal(env, dt, name, th, capa);
+ if (rc == -ENODATA)
+ rc = 0;
+ } else {
+ rc = lod_xattr_set_internal(env, dt, buf, name, fl, th, capa);
+ if (rc != 0)
+ RETURN(rc);
+ }
+
+ /* Update default stripe cache */
+ if (l->ldo_dir_stripe == NULL) {
+ OBD_ALLOC_PTR(l->ldo_dir_stripe);
+ if (l->ldo_dir_stripe == NULL)
+ RETURN(-ENOMEM);
+ }
+
+ l->ldo_dir_striping_cached = 0;
+ l->ldo_dir_def_striping_set = 1;
+ l->ldo_dir_def_stripenr = le32_to_cpu(lum->lum_stripe_count);
+
+ RETURN(rc);
+}
+
+static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
+ const struct lu_buf *buf, const char *name,
+ int fl, struct thandle *th,
+ struct lustre_capa *capa)
+{
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lu_attr *attr = &info->lti_attr;
+ struct dt_object_format *dof = &info->lti_format;
+ struct lu_buf lmv_buf;
+ struct lu_buf slave_lmv_buf;
+ struct lmv_mds_md_v1 *lmm;
+ struct lmv_mds_md_v1 *slave_lmm = NULL;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ int i;
+ int rc;
+ ENTRY;
+
+ if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
+ RETURN(-ENOTDIR);
+
+ /* The stripes are supposed to be allocated in declare phase,
+ * if there are no stripes being allocated, it will skip */
+ if (lo->ldo_stripenr == 0)
+ RETURN(0);
+
+ rc = dt_attr_get(env, dt_object_child(dt), attr, BYPASS_CAPA);
+ if (rc != 0)
+ RETURN(rc);
+
+ attr->la_valid = LA_TYPE | LA_MODE;
+ dof->dof_type = DFT_DIR;
+
+ rc = lod_prep_lmv_md(env, dt, &lmv_buf);
+ if (rc != 0)
+ RETURN(rc);
+ lmm = lmv_buf.lb_buf;
+
+ OBD_ALLOC_PTR(slave_lmm);
+ if (slave_lmm == NULL)
+ RETURN(-ENOMEM);
+
+ lod_prep_slave_lmv_md(slave_lmm, lmm);
+ slave_lmv_buf.lb_buf = slave_lmm;
+ slave_lmv_buf.lb_len = sizeof(*slave_lmm);
+
+ rec->rec_type = S_IFDIR;
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ struct dt_object *dto;
+ char *stripe_name = info->lti_key;
+ struct lu_name *sname;
+ struct linkea_data ldata = { 0 };
+ struct lu_buf linkea_buf;
+
+ dto = lo->ldo_stripe[i];
+ dt_write_lock(env, dto, MOR_TGT_CHILD);
+ rc = dt_create(env, dto, attr, NULL, dof, th);
+ dt_write_unlock(env, dto);
+ if (rc != 0)
+ RETURN(rc);
+
+ rec->rec_fid = lu_object_fid(&dto->do_lu);
+ rc = dt_insert(env, dto, (const struct dt_rec *)rec,
+ (const struct dt_key *)dot, th, capa, 0);
+ if (rc != 0)
+ RETURN(rc);
+
+ rec->rec_fid = lu_object_fid(&dt->do_lu);
+ rc = dt_insert(env, dto, (struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th, capa, 0);
+ if (rc != 0)
+ RETURN(rc);
+
+ if (lo->ldo_striping_cached &&
+ !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
+ lo->ldo_def_stripenr,
+ lo->ldo_def_stripe_offset)) {
+ struct lov_user_md_v3 *v3;
+
+ /* sigh, lti_ea_store has been used for lmv_buf,
+ * so we have to allocate buffer for default
+ * stripe EA */
+ OBD_ALLOC_PTR(v3);
+ if (v3 == NULL)
+ GOTO(out, rc);
+
+ memset(v3, 0, sizeof(*v3));
+ v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
+ v3->lmm_stripe_count =
+ cpu_to_le16(lo->ldo_def_stripenr);
+ v3->lmm_stripe_offset =
+ cpu_to_le16(lo->ldo_def_stripe_offset);
+ v3->lmm_stripe_size =
+ cpu_to_le32(lo->ldo_def_stripe_size);
+ if (lo->ldo_pool != NULL)
+ strlcpy(v3->lmm_pool_name, lo->ldo_pool,
+ sizeof(v3->lmm_pool_name));
+
+ info->lti_buf.lb_buf = v3;
+ info->lti_buf.lb_len = sizeof(*v3);
+ rc = dt_xattr_set(env, dto, &info->lti_buf,
+ XATTR_NAME_LOV, 0, th, capa);
+ OBD_FREE_PTR(v3);
+ if (rc != 0)
+ GOTO(out, rc);
+ }
+
+ slave_lmm->lmv_master_mdt_index = cpu_to_le32(i);
+ rc = dt_xattr_set(env, dto, &slave_lmv_buf, XATTR_NAME_LMV,
+ fl, th, capa);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
+ PFID(lu_object_fid(&dto->do_lu)), i);
+
+ sname = lod_name_get(env, stripe_name, strlen(stripe_name));
+ rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
+ if (rc != 0)
+ GOTO(out, rc);
+
+ linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
+ linkea_buf.lb_len = ldata.ld_leh->leh_len;
+ rc = dt_xattr_set(env, dto, &linkea_buf, XATTR_NAME_LINK,
+ 0, th, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ rec->rec_fid = lu_object_fid(&dto->do_lu);
+ rc = dt_insert(env, dt_object_child(dt),
+ (const struct dt_rec *)rec,
+ (const struct dt_key *)stripe_name, th, capa, 0);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ rc = dt_ref_add(env, dt_object_child(dt), th);
+ if (rc != 0)
+ GOTO(out, rc);
+ }
+
+ rc = dt_xattr_set(env, dt_object_child(dt), &lmv_buf, XATTR_NAME_LMV,
+ fl, th, capa);
+
+out:
+ if (slave_lmm != NULL)
+ OBD_FREE_PTR(slave_lmm);
+
+ RETURN(rc);
+}
+
+int lod_dir_striping_create_internal(const struct lu_env *env,
+ struct dt_object *dt,
+ struct lu_attr *attr,
+ struct dt_object_format *dof,
+ struct thandle *th,
+ bool declare)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_object *lo = lod_dt_obj(dt);
+ int rc;
+ ENTRY;
+
+ if (!LMVEA_DELETE_VALUES(lo->ldo_stripenr,
+ lo->ldo_dir_stripe_offset)) {
+ struct lmv_user_md_v1 *v1 = info->lti_ea_store;
+ int stripe_count = lo->ldo_stripenr;
+
+ if (info->lti_ea_store_size < sizeof(*v1)) {
+ rc = lod_ea_store_resize(info, sizeof(*v1));
+ if (rc != 0)
+ RETURN(rc);
+ v1 = info->lti_ea_store;
+ }
+
+ memset(v1, 0, sizeof(*v1));
+ v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
+ v1->lum_stripe_count = cpu_to_le32(stripe_count);
+ v1->lum_stripe_offset =
+ cpu_to_le32(lo->ldo_dir_stripe_offset);
+
+ info->lti_buf.lb_buf = v1;
+ info->lti_buf.lb_len = sizeof(*v1);
+
+ if (declare)
+ rc = lod_declare_xattr_set_lmv(env, dt, attr,
+ &info->lti_buf, dof, th);
+ else
+ rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
+ XATTR_NAME_LMV, 0, th,
+ BYPASS_CAPA);
+ if (rc != 0)
+ RETURN(rc);
+ }
+
+ /* Transfer default LMV striping from the parent */
+ if (lo->ldo_dir_striping_cached &&
+ !LMVEA_DELETE_VALUES(lo->ldo_dir_def_stripenr,
+ lo->ldo_dir_def_stripe_offset)) {
+ struct lmv_user_md_v1 *v1 = info->lti_ea_store;
+ int def_stripe_count = lo->ldo_dir_def_stripenr;
+
+ if (info->lti_ea_store_size < sizeof(*v1)) {
+ rc = lod_ea_store_resize(info, sizeof(*v1));
+ if (rc != 0)
+ RETURN(rc);
+ v1 = info->lti_ea_store;
+ }
+
+ memset(v1, 0, sizeof(*v1));
+ v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
+ v1->lum_stripe_count = cpu_to_le32(def_stripe_count);
+ v1->lum_stripe_offset =
+ cpu_to_le32(lo->ldo_dir_def_stripe_offset);
+ v1->lum_hash_type =
+ cpu_to_le32(lo->ldo_dir_def_hash_type);
+
+ info->lti_buf.lb_buf = v1;
+ info->lti_buf.lb_len = sizeof(*v1);
+ if (declare)
+ rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
+ XATTR_NAME_DEFAULT_LMV,
+ 0, th);
+ else
+ rc = lod_xattr_set_default_lmv_on_dir(env, dt,
+ &info->lti_buf,
+ XATTR_NAME_DEFAULT_LMV, 0,
+ th, BYPASS_CAPA);
+ if (rc != 0)
+ RETURN(rc);
+ }