static int out_create(struct mdt_thread_info *info)
{
struct update *update = info->mti_u.update.mti_update;
struct dt_object *obj = info->mti_u.update.mti_dt_object;
struct dt_object_format *dof = &info->mti_u.update.mti_update_dof;
struct obdo *lobdo = &info->mti_u.update.mti_obdo;
struct lu_attr *attr = &info->mti_attr.ma_attr;
struct lu_fid *fid = NULL;
struct obdo *wobdo;
int size;
int rc;
ENTRY;
wobdo = update_param_buf(update, 0, &size);
if (wobdo == NULL || size != sizeof(*wobdo)) {
CERROR("%s: obdo is NULL, invalid RPC: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
obdo_le_to_cpu(wobdo, wobdo);
lustre_get_wire_obdo(lobdo, wobdo);
la_from_obdo(attr, lobdo, lobdo->o_valid);
dof->dof_type = dt_mode_to_dft(attr->la_mode);
if (S_ISDIR(attr->la_mode)) {
int size;
fid = update_param_buf(update, 1, &size);
if (fid == NULL || size != sizeof(*fid)) {
CERROR("%s: invalid fid: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
fid_le_to_cpu(fid, fid);
if (!fid_is_sane(fid)) {
CERROR("%s: invalid fid "DFID": rc = %d\n",
mdt_obd_name(info->mti_mdt),
PFID(fid), -EPROTO);
RETURN(err_serious(-EPROTO));
}
}
rc = out_tx_create(info, obj, attr, fid, dof, &info->mti_handle,
info->mti_u.update.mti_update_reply,
info->mti_u.update.mti_update_reply_index);
RETURN(rc);
}
static int out_attr_set(struct mdt_thread_info *info)
{
struct update *update = info->mti_u.update.mti_update;
struct lu_attr *attr = &info->mti_attr.ma_attr;
struct dt_object *obj = info->mti_u.update.mti_dt_object;
struct obdo *lobdo = &info->mti_u.update.mti_obdo;
struct obdo *wobdo;
int size;
int rc;
ENTRY;
wobdo = update_param_buf(update, 0, &size);
if (wobdo == NULL || size != sizeof(*wobdo)) {
CERROR("%s: empty obdo in the update: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
attr->la_valid = 0;
attr->la_valid = 0;
obdo_le_to_cpu(wobdo, wobdo);
lustre_get_wire_obdo(lobdo, wobdo);
la_from_obdo(attr, lobdo, lobdo->o_valid);
rc = out_tx_attr_set(info, obj, attr, &info->mti_handle,
info->mti_u.update.mti_update_reply,
info->mti_u.update.mti_update_reply_index);
RETURN(rc);
}
static int out_tx_index_insert_undo(struct mdt_thread_info *info,
struct thandle *th, struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
CDEBUG(D_OTHER, "index delete "DFID" name: %s\n",
PFID(lu_object_fid(&arg->object->do_lu)),
(char *)arg->u.insert.key);
if (dt_try_as_dir(info->mti_env, dt_obj) == 0) {
CERROR("%s: "DFID" is not directory: rc = %d\n",
mdt_obd_name(info->mti_mdt),
PFID(lu_object_fid(&dt_obj->do_lu)), -ENOTDIR);
return -ENOTDIR;
}
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
rc = dt_delete(info->mti_env, dt_obj, arg->u.insert.key, th, NULL);
dt_write_unlock(info->mti_env, dt_obj);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
static int out_xattr_get(struct mdt_thread_info *info)
{
struct update *update = info->mti_u.update.mti_update;
const struct lu_env *env = info->mti_env;
struct lu_buf *lbuf = &info->mti_buf;
struct update_reply *reply = info->mti_u.update.mti_update_reply;
struct dt_object *obj = info->mti_u.update.mti_dt_object;
char *name;
void *ptr;
int rc;
ENTRY;
name = (char *)update_param_buf(update, 0, NULL);
if (name == NULL) {
CERROR("%s: empty name for xattr get: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
ptr = update_get_buf_internal(reply, 0, NULL);
LASSERT(ptr != NULL);
/* The first 4 bytes(int) are used to store the result */
lbuf->lb_buf = (char *)ptr + sizeof(int);
lbuf->lb_len = UPDATE_BUFFER_SIZE - sizeof(struct update_reply);
dt_read_lock(env, obj, MOR_TGT_CHILD);
rc = dt_xattr_get(env, obj, lbuf, name, NULL);
dt_read_unlock(env, obj);
if (rc < 0) {
lbuf->lb_len = 0;
GOTO(out, rc);
}
if (rc == 0) {
lbuf->lb_len = 0;
GOTO(out, rc = -ENOENT);
}
lbuf->lb_len = rc;
rc = 0;
CDEBUG(D_INFO, "%s: "DFID" get xattr %s len %d\n",
mdt_obd_name(info->mti_mdt), PFID(lu_object_fid(&obj->do_lu)),
name, (int)lbuf->lb_len);
out:
*(int *)ptr = rc;
reply->ur_lens[0] = lbuf->lb_len + sizeof(int);
RETURN(rc);
}
static int out_xattr_set(struct mdt_thread_info *info)
{
struct update *update = info->mti_u.update.mti_update;
struct dt_object *obj = info->mti_u.update.mti_dt_object;
struct lu_buf *lbuf = &info->mti_buf;
char *name;
char *buf;
char *tmp;
int buf_len = 0;
int flag;
int rc;
ENTRY;
name = update_param_buf(update, 0, NULL);
if (name == NULL) {
CERROR("%s: empty name for xattr set: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
buf = (char *)update_param_buf(update, 1, &buf_len);
if (buf == NULL || buf_len == 0) {
CERROR("%s: empty buf for xattr set: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
lbuf->lb_buf = buf;
lbuf->lb_len = buf_len;
tmp = (char *)update_param_buf(update, 2, NULL);
if (tmp == NULL) {
CERROR("%s: empty flag for xattr set: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
flag = le32_to_cpu(*(int *)tmp);
rc = out_tx_xattr_set(info, obj, lbuf, name, flag, &info->mti_handle,
info->mti_u.update.mti_update_reply,
info->mti_u.update.mti_update_reply_index);
RETURN(rc);
}
static int out_tx_attr_set_undo(struct mdt_thread_info *info,
struct thandle *th, struct tx_arg *arg)
{
CERROR("%s: attr set undo "DFID" unimplemented yet!: rc = %d\n",
mdt_obd_name(info->mti_mdt),
PFID(lu_object_fid(&arg->object->do_lu)), -ENOTSUPP);
return -ENOTSUPP;
}
static int out_index_insert(struct mdt_thread_info *info)
{
struct update *update = info->mti_u.update.mti_update;
struct dt_object *obj = info->mti_u.update.mti_dt_object;
struct lu_fid *fid;
char *name;
int rc = 0;
int size;
ENTRY;
name = (char *)update_param_buf(update, 0, NULL);
if (name == NULL) {
CERROR("%s: empty name for index insert: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
fid = (struct lu_fid *)update_param_buf(update, 1, &size);
if (fid == NULL || size != sizeof(*fid)) {
CERROR("%s: invalid fid: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
fid_le_to_cpu(fid, fid);
if (!fid_is_sane(fid)) {
CERROR("%s: invalid FID "DFID": rc = %d\n",
mdt_obd_name(info->mti_mdt), PFID(fid),
-EPROTO);
RETURN(err_serious(-EPROTO));
}
rc = out_tx_index_insert(info, obj, name, fid, &info->mti_handle,
info->mti_u.update.mti_update_reply,
info->mti_u.update.mti_update_reply_index);
RETURN(rc);
}
static int out_index_lookup(struct mdt_thread_info *info)
{
struct update *update = info->mti_u.update.mti_update;
const struct lu_env *env = info->mti_env;
struct dt_object *obj = info->mti_u.update.mti_dt_object;
char *name;
int rc;
ENTRY;
if (!lu_object_exists(&obj->do_lu))
RETURN(-ENOENT);
name = (char *)update_param_buf(update, 0, NULL);
if (name == NULL) {
CERROR("%s: empty name for lookup: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
dt_read_lock(env, obj, MOR_TGT_CHILD);
if (!dt_try_as_dir(env, obj))
GOTO(out_unlock, rc = -ENOTDIR);
rc = dt_lookup(env, obj, (struct dt_rec *)&info->mti_tmp_fid1,
(struct dt_key *)name, NULL);
if (rc < 0)
GOTO(out_unlock, rc);
if (rc == 0)
rc += 1;
CDEBUG(D_INFO, "lookup "DFID" %s get "DFID" rc %d\n",
PFID(lu_object_fid(&obj->do_lu)), name,
PFID(&info->mti_tmp_fid1), rc);
fid_cpu_to_le(&info->mti_tmp_fid1, &info->mti_tmp_fid1);
out_unlock:
dt_read_unlock(env, obj);
update_insert_reply(info->mti_u.update.mti_update_reply,
&info->mti_tmp_fid1, sizeof(info->mti_tmp_fid1),
0, rc);
RETURN(rc);
}
/**
* All of the xxx_undo will be used once execution failed,
* But because all of the required resource has been reserved in
* declare phase, i.e. if declare succeed, it should make sure
* the following executing phase succeed in anyway, so these undo
* should be useless for most of the time in Phase I
**/
int out_tx_create_undo(struct mdt_thread_info *info, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
rc = dt_destroy(info->mti_env, dt_obj, th);
dt_write_unlock(info->mti_env, dt_obj);
/* we don't like double failures */
if (rc != 0)
CERROR("%s: undo failure, we are doomed!: rc = %d\n",
mdt_obd_name(info->mti_mdt), rc);
return rc;
}
int out_tx_end(struct mdt_thread_info *info, struct thandle_exec_args *th)
{
struct thandle_exec_args *_th = &info->mti_handle;
int i = 0, rc;
LASSERT(th == _th);
LASSERT(th->ta_dev);
LASSERT(th->ta_handle);
if (th->ta_err != 0 || th->ta_argno == 0)
GOTO(stop, rc = th->ta_err);
rc = out_trans_start(info->mti_env, th->ta_dev, th->ta_handle);
if (unlikely(rc))
GOTO(stop, rc);
for (i = 0; i < th->ta_argno; i++) {
rc = th->ta_args[i].exec_fn(info, th->ta_handle,
&th->ta_args[i]);
if (unlikely(rc)) {
CDEBUG(D_INFO, "error during execution of #%u from"
" %s:%d: rc = %d\n", i, th->ta_args[i].file,
th->ta_args[i].line, rc);
while (--i >= 0) {
LASSERTF(th->ta_args[i].undo_fn != NULL,
"can't undo changes, hope for failover!\n");
th->ta_args[i].undo_fn(info, th->ta_handle,
&th->ta_args[i]);
}
break;
}
}
stop:
CDEBUG(D_INFO, "%s: executed %u/%u: rc = %d\n",
mdt_obd_name(info->mti_mdt), i, th->ta_argno, rc);
out_trans_stop(info->mti_env, th, rc);
th->ta_handle = NULL;
th->ta_argno = 0;
th->ta_err = 0;
RETURN(rc);
}
static int out_index_delete(struct mdt_thread_info *info)
{
struct update *update = info->mti_u.update.mti_update;
struct dt_object *obj = info->mti_u.update.mti_dt_object;
char *name;
int rc = 0;
if (!lu_object_exists(&obj->do_lu))
RETURN(-ENOENT);
name = (char *)update_param_buf(update, 0, NULL);
if (name == NULL) {
CERROR("%s: empty name for index delete: rc = %d\n",
mdt_obd_name(info->mti_mdt), -EPROTO);
RETURN(err_serious(-EPROTO));
}
rc = out_tx_index_delete(info, obj, name, &info->mti_handle,
info->mti_u.update.mti_update_reply,
info->mti_u.update.mti_update_reply_index);
RETURN(rc);
}
static int mdt_reint_create(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct ptlrpc_request *req = mdt_info_req(info);
int rc;
ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
RETURN(err_serious(-ESTALE));
if (info->mti_dlm_req)
ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
LASSERT(info->mti_rr.rr_namelen > 0);
switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
mdt_counter_incr(req, LPROC_MDT_MKDIR);
break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
/* Special file should stay on the same node as parent. */
mdt_counter_incr(req, LPROC_MDT_MKNOD);
break;
default:
CERROR("%s: Unsupported mode %o\n",
mdt_obd_name(info->mti_mdt),
info->mti_attr.ma_attr.la_mode);
RETURN(err_serious(-EOPNOTSUPP));
}
rc = mdt_md_create(info);
RETURN(rc);
}
/**
* Retrieve undergoing HSM requests for the fid provided in RPC body.
* Current requests are read from coordinator states.
*
* This is MDS_HSM_ACTION RPC handler.
*/
int mdt_hsm_action(struct tgt_session_info *tsi)
{
struct mdt_thread_info *info;
struct hsm_current_action *hca;
struct hsm_action_list *hal = NULL;
struct hsm_action_item *hai;
int hal_size;
int rc;
ENTRY;
hca = req_capsule_server_get(tsi->tsi_pill,
&RMF_MDS_HSM_CURRENT_ACTION);
if (hca == NULL)
RETURN(err_serious(-EPROTO));
if (tsi->tsi_mdt_body == NULL)
RETURN(-EPROTO);
info = tsi2mdt_info(tsi);
/* Only valid if client is remote */
rc = mdt_init_ucred(info, (struct mdt_body *)info->mti_body);
if (rc)
GOTO(out, rc = err_serious(rc));
/* Coordinator information */
hal_size = sizeof(*hal) +
cfs_size_round(MTI_NAME_MAXLEN) /* fsname */ +
cfs_size_round(sizeof(*hai));
MDT_HSM_ALLOC(hal, hal_size);
if (hal == NULL)
GOTO(out_ucred, rc = -ENOMEM);
hal->hal_version = HAL_VERSION;
hal->hal_archive_id = 0;
hal->hal_flags = 0;
obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(info->mti_mdt),
MTI_NAME_MAXLEN);
hal->hal_count = 1;
hai = hai_first(hal);
hai->hai_action = HSMA_NONE;
hai->hai_cookie = 0;
hai->hai_gid = 0;
hai->hai_fid = info->mti_body->mbo_fid1;
hai->hai_len = sizeof(*hai);
rc = mdt_hsm_get_actions(info, hal);
if (rc)
GOTO(out_free, rc);
/* cookie is used to give back request status */
if (hai->hai_cookie == 0)
hca->hca_state = HPS_WAITING;
else
hca->hca_state = HPS_RUNNING;
switch (hai->hai_action) {
case HSMA_NONE:
hca->hca_action = HUA_NONE;
break;
case HSMA_ARCHIVE:
hca->hca_action = HUA_ARCHIVE;
break;
case HSMA_RESTORE:
hca->hca_action = HUA_RESTORE;
break;
case HSMA_REMOVE:
hca->hca_action = HUA_REMOVE;
break;
case HSMA_CANCEL:
hca->hca_action = HUA_CANCEL;
break;
default:
hca->hca_action = HUA_NONE;
CERROR("%s: Unknown hsm action: %d on "DFID"\n",
mdt_obd_name(info->mti_mdt),
hai->hai_action, PFID(&hai->hai_fid));
break;
}
hca->hca_location = hai->hai_extent;
EXIT;
out_free:
MDT_HSM_FREE(hal, hal_size);
out_ucred:
mdt_exit_ucred(info);
out:
mdt_thread_info_fini(info);
return rc;
}
static int mdt_lvbo_fill(struct ldlm_lock *lock, void *lvb, int lvblen)
{
struct lu_env env;
struct mdt_thread_info *info;
struct mdt_device *mdt;
struct lu_fid *fid;
struct mdt_object *obj = NULL;
struct md_object *child = NULL;
int rc;
ENTRY;
mdt = ldlm_lock_to_ns(lock)->ns_lvbp;
if (IS_LQUOTA_RES(lock->l_resource)) {
if (mdt->mdt_qmt_dev == NULL)
RETURN(0);
/* call lvbo fill function of quota master */
rc = qmt_hdls.qmth_lvbo_fill(mdt->mdt_qmt_dev, lock, lvb,
lvblen);
RETURN(rc);
}
if (!ldlm_has_layout(lock))
RETURN(0);
/* layout lock will be granted to client, fill in lvb with layout */
/* XXX create an env to talk to mdt stack. We should get this env from
* ptlrpc_thread->t_env. */
rc = lu_env_init(&env, LCT_MD_THREAD);
/* Likely ENOMEM */
if (rc)
RETURN(rc);
info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
/* Likely ENOMEM */
if (info == NULL)
GOTO(out, rc = -ENOMEM);
memset(info, 0, sizeof *info);
info->mti_env = &env;
info->mti_exp = lock->l_export;
info->mti_mdt = mdt;
/* XXX get fid by resource id. why don't include fid in ldlm_resource */
fid = &info->mti_tmp_fid2;
fid_extract_from_res_name(fid, &lock->l_resource->lr_name);
obj = mdt_object_find(&env, info->mti_mdt, fid);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
if (!mdt_object_exists(obj) || mdt_object_remote(obj))
GOTO(out, rc = -ENOENT);
child = mdt_object_child(obj);
/* get the length of lsm */
rc = mo_xattr_get(&env, child, &LU_BUF_NULL, XATTR_NAME_LOV);
if (rc < 0)
GOTO(out, rc);
if (rc > 0) {
struct lu_buf *lmm = NULL;
if (lvblen < rc) {
CERROR("%s: expected %d actual %d.\n",
mdt_obd_name(mdt), rc, lvblen);
GOTO(out, rc = -ERANGE);
}
lmm = &info->mti_buf;
lmm->lb_buf = lvb;
lmm->lb_len = rc;
rc = mo_xattr_get(&env, child, lmm, XATTR_NAME_LOV);
if (rc < 0)
GOTO(out, rc);
}
out:
if (obj != NULL && !IS_ERR(obj))
mdt_object_put(&env, obj);
lu_env_fini(&env);
RETURN(rc < 0 ? 0 : rc);
}
/* return EADATA length to the caller. negative value means error */
static int mdt_getxattr_pack_reply(struct mdt_thread_info * info)
{
struct req_capsule *pill = info->mti_pill;
struct ptlrpc_request *req = mdt_info_req(info);
const char *xattr_name;
u64 valid;
static const char user_string[] = "user.";
int size;
int rc = 0;
int rc2;
ENTRY;
valid = info->mti_body->mbo_valid & (OBD_MD_FLXATTR | OBD_MD_FLXATTRLS);
/* Determine how many bytes we need */
if (valid == OBD_MD_FLXATTR) {
xattr_name = req_capsule_client_get(pill, &RMF_NAME);
if (!xattr_name)
RETURN(-EFAULT);
if (!(exp_connect_flags(req->rq_export) & OBD_CONNECT_XATTR) &&
!strncmp(xattr_name, user_string, sizeof(user_string) - 1))
RETURN(-EOPNOTSUPP);
size = mo_xattr_get(info->mti_env,
mdt_object_child(info->mti_object),
&LU_BUF_NULL, xattr_name);
if (size == -ENODATA) {
/* XXX: Some client code will not handle -ENODATA
* for XATTR_NAME_LOV (trusted.lov) properly. */
if (strcmp(xattr_name, XATTR_NAME_LOV) == 0)
rc = 0;
else
rc = -ENODATA;
size = 0;
}
} else if (valid == OBD_MD_FLXATTRLS) {
xattr_name = "list";
size = mo_xattr_list(info->mti_env,
mdt_object_child(info->mti_object),
&LU_BUF_NULL);
} else if (valid == OBD_MD_FLXATTRALL) {
xattr_name = "all";
/* N.B. eadatasize = 0 is not valid for FLXATTRALL */
/* We could calculate accurate sizes, but this would
* introduce a lot of overhead, let's do it later... */
size = info->mti_body->mbo_eadatasize;
req_capsule_set_size(pill, &RMF_EAVALS, RCL_SERVER, size);
req_capsule_set_size(pill, &RMF_EAVALS_LENS, RCL_SERVER, size);
} else {
CDEBUG(D_INFO, "Valid bits: %#llx\n",
info->mti_body->mbo_valid);
RETURN(-EINVAL);
}
if (size < 0) {
if (size != -EOPNOTSUPP && size != -ENOENT)
CERROR("%s: error geting EA size for '%s': rc = %d\n",
mdt_obd_name(info->mti_mdt), xattr_name, size);
RETURN(size);
}
if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
req_capsule_set_size(pill, &RMF_EADATA, RCL_SERVER,
info->mti_body->mbo_eadatasize == 0 ? 0 : size);
rc2 = req_capsule_server_pack(pill);
if (rc2 < 0)
RETURN(rc2);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETXATTR_PACK))
RETURN(-ENOMEM);
RETURN(rc < 0 ? rc : size);
}
int mdt_reint_setxattr(struct mdt_thread_info *info,
struct mdt_lock_handle *unused)
{
struct ptlrpc_request *req = mdt_info_req(info);
struct mdt_lock_handle *lh;
const struct lu_env *env = info->mti_env;
struct lu_buf *buf = &info->mti_buf;
struct mdt_reint_record *rr = &info->mti_rr;
struct md_attr *ma = &info->mti_attr;
struct lu_attr *attr = &info->mti_attr.ma_attr;
struct mdt_object *obj;
struct md_object *child;
__u64 valid = attr->la_valid;
const char *xattr_name = rr->rr_name.ln_name;
int xattr_len = rr->rr_eadatalen;
__u64 lockpart = MDS_INODELOCK_UPDATE;
int rc;
ENTRY;
CDEBUG(D_INODE, "setxattr for "DFID": %s %s\n", PFID(rr->rr_fid1),
valid & OBD_MD_FLXATTR ? "set" : "remove", xattr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SETXATTR))
RETURN(err_serious(-ENOMEM));
rc = mdt_init_ucred_reint(info);
if (rc != 0)
RETURN(rc);
if (strncmp(xattr_name, XATTR_USER_PREFIX,
sizeof(XATTR_USER_PREFIX) - 1) == 0) {
if (!(exp_connect_flags(req->rq_export) & OBD_CONNECT_XATTR))
GOTO(out, rc = -EOPNOTSUPP);
} else if (strncmp(xattr_name, XATTR_TRUSTED_PREFIX,
sizeof(XATTR_TRUSTED_PREFIX) - 1) == 0) {
/* setxattr(LMV) with lum is used to shrink dir layout */
if (strcmp(xattr_name, XATTR_NAME_LMV) == 0) {
__u32 *magic = rr->rr_eadata;
/* we don't let to remove LMV? */
if (!rr->rr_eadata)
GOTO(out, rc = 0);
if (le32_to_cpu(*magic) == LMV_USER_MAGIC ||
le32_to_cpu(*magic) == LMV_USER_MAGIC_SPECIFIC) {
rc = mdt_dir_layout_shrink(info);
GOTO(out, rc);
}
}
if (!md_capable(mdt_ucred(info), CFS_CAP_SYS_ADMIN))
GOTO(out, rc = -EPERM);
if (strcmp(xattr_name, XATTR_NAME_LOV) == 0 ||
strcmp(xattr_name, XATTR_NAME_LMA) == 0 ||
strcmp(xattr_name, XATTR_NAME_LMV) == 0 ||
strcmp(xattr_name, XATTR_NAME_LINK) == 0 ||
strcmp(xattr_name, XATTR_NAME_FID) == 0 ||
strcmp(xattr_name, XATTR_NAME_VERSION) == 0 ||
strcmp(xattr_name, XATTR_NAME_SOM) == 0 ||
strcmp(xattr_name, XATTR_NAME_HSM) == 0 ||
strcmp(xattr_name, XATTR_NAME_LFSCK_NAMESPACE) == 0)
GOTO(out, rc = 0);
} else if ((valid & OBD_MD_FLXATTR) &&
(strcmp(xattr_name, XATTR_NAME_ACL_ACCESS) == 0 ||
strcmp(xattr_name, XATTR_NAME_ACL_DEFAULT) == 0)) {
rc = mdt_nodemap_map_acl(info, rr->rr_eadata, xattr_len,
xattr_name, NODEMAP_CLIENT_TO_FS);
if (rc < 0)
GOTO(out, rc);
/* ACLs were mapped out, return an error so the user knows */
if (rc != xattr_len)
GOTO(out, rc = -EPERM);
} else if ((strlen(xattr_name) > strlen(XATTR_LUSTRE_LOV) + 1) &&
strncmp(xattr_name, XATTR_LUSTRE_LOV,
strlen(XATTR_LUSTRE_LOV)) == 0) {
if (strncmp(xattr_name, XATTR_LUSTRE_LOV".add",
strlen(XATTR_LUSTRE_LOV".add")) &&
strncmp(xattr_name, XATTR_LUSTRE_LOV".set",
strlen(XATTR_LUSTRE_LOV".set")) &&
strncmp(xattr_name, XATTR_LUSTRE_LOV".del",
strlen(XATTR_LUSTRE_LOV".del"))) {
CERROR("%s: invalid xattr name: %s\n",
mdt_obd_name(info->mti_mdt), xattr_name);
GOTO(out, rc = -EINVAL);
}
lockpart |= MDS_INODELOCK_LAYOUT;
}
/* Revoke all clients' lookup lock, since the access
* permissions for this inode is changed when ACL_ACCESS is
* set. This isn't needed for ACL_DEFAULT, since that does
* not change the access permissions of this inode, nor any
* other existing inodes. It is setting the ACLs inherited
* by new directories/files at create time. */
/* We need revoke both LOOKUP|PERM lock here, see mdt_attr_set. */
if (!strcmp(xattr_name, XATTR_NAME_ACL_ACCESS))
lockpart |= MDS_INODELOCK_PERM | MDS_INODELOCK_LOOKUP;
/* We need to take the lock on behalf of old clients so that newer
* clients flush their xattr caches */
else
lockpart |= MDS_INODELOCK_XATTR;
lh = &info->mti_lh[MDT_LH_PARENT];
/* ACLs were sent to clients under LCK_CR locks, so taking LCK_EX
* to cancel them. */
mdt_lock_reg_init(lh, LCK_EX);
obj = mdt_object_find_lock(info, rr->rr_fid1, lh, lockpart);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
tgt_vbr_obj_set(env, mdt_obj2dt(obj));
rc = mdt_version_get_check_save(info, obj, 0);
if (rc)
GOTO(out_unlock, rc);
if (unlikely(!(valid & OBD_MD_FLCTIME))) {
/* This isn't strictly an error, but all current clients
* should set OBD_MD_FLCTIME when setting attributes. */
CWARN("%s: client miss to set OBD_MD_FLCTIME when "
"setxattr %s: [object "DFID"] [valid %llu]\n",
mdt_obd_name(info->mti_mdt), xattr_name,
PFID(rr->rr_fid1), valid);
attr->la_ctime = ktime_get_real_seconds();
}
attr->la_valid = LA_CTIME;
child = mdt_object_child(obj);
if (valid & OBD_MD_FLXATTR) {
int flags = 0;
if (attr->la_flags & XATTR_REPLACE)
flags |= LU_XATTR_REPLACE;
if (attr->la_flags & XATTR_CREATE)
flags |= LU_XATTR_CREATE;
mdt_fail_write(env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_SETXATTR_WRITE);
buf->lb_buf = rr->rr_eadata;
buf->lb_len = xattr_len;
rc = mo_xattr_set(env, child, buf, xattr_name, flags);
/* update ctime after xattr changed */
if (rc == 0) {
ma->ma_attr_flags |= MDS_PERM_BYPASS;
mo_attr_set(env, child, ma);
}
} else if (valid & OBD_MD_FLXATTRRM) {
rc = mo_xattr_del(env, child, xattr_name);
/* update ctime after xattr changed */
if (rc == 0) {
ma->ma_attr_flags |= MDS_PERM_BYPASS;
mo_attr_set(env, child, ma);
}
} else {
CDEBUG(D_INFO, "valid bits: %#llx\n", valid);
rc = -EINVAL;
}
if (rc == 0)
mdt_counter_incr(req, LPROC_MDT_SETXATTR);
EXIT;
out_unlock:
mdt_object_unlock_put(info, obj, lh, rc);
out:
mdt_exit_ucred(info);
return rc;
}
/* shrink dir layout after migration */
static int mdt_dir_layout_shrink(struct mdt_thread_info *info)
{
const struct lu_env *env = info->mti_env;
struct mdt_device *mdt = info->mti_mdt;
struct lu_ucred *uc = mdt_ucred(info);
struct mdt_reint_record *rr = &info->mti_rr;
struct lmv_user_md *lmu = rr->rr_eadata;
__u32 lum_stripe_count = lmu->lum_stripe_count;
struct lu_buf *buf = &info->mti_buf;
struct lmv_mds_md_v1 *lmv;
struct md_attr *ma = &info->mti_attr;
struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
struct mdt_object *pobj = NULL;
struct mdt_object *obj;
struct mdt_lock_handle *lhp = NULL;
struct mdt_lock_handle *lhc;
int rc;
ENTRY;
if (!mdt->mdt_enable_dir_migration)
RETURN(-EPERM);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN) &&
uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
mdt->mdt_enable_remote_dir_gid != -1)
RETURN(-EPERM);
/* mti_big_lmm is used to save LMV, but it may be uninitialized. */
if (unlikely(!info->mti_big_lmm)) {
info->mti_big_lmmsize = lmv_mds_md_size(64, LMV_MAGIC);
OBD_ALLOC(info->mti_big_lmm, info->mti_big_lmmsize);
if (!info->mti_big_lmm)
RETURN(-ENOMEM);
}
obj = mdt_object_find(env, mdt, rr->rr_fid1);
if (IS_ERR(obj))
RETURN(PTR_ERR(obj));
/* get parent from PFID */
rc = mdt_attr_get_pfid(info, obj, &ma->ma_pfid);
if (rc)
GOTO(put_obj, rc);
pobj = mdt_object_find(env, mdt, &ma->ma_pfid);
if (IS_ERR(pobj))
GOTO(put_obj, rc = PTR_ERR(pobj));
/* revoke object remote LOOKUP lock */
if (mdt_object_remote(pobj)) {
rc = mdt_revoke_remote_lookup_lock(info, pobj, obj);
if (rc)
GOTO(put_pobj, rc);
}
/*
* lock parent if dir will be shrunk to 1 stripe, because dir will be
* converted to normal directory, as will change dir fid and update
* namespace of parent.
*/
lhp = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_reg_init(lhp, LCK_PW);
if (le32_to_cpu(lmu->lum_stripe_count) < 2) {
rc = mdt_reint_object_lock(info, pobj, lhp,
MDS_INODELOCK_UPDATE, true);
if (rc)
GOTO(put_pobj, rc);
}
/* lock object */
lhc = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(lhc, LCK_EX);
rc = mdt_reint_striped_lock(info, obj, lhc, MDS_INODELOCK_FULL, einfo,
true);
if (rc)
GOTO(unlock_pobj, rc);
ma->ma_lmv = info->mti_big_lmm;
ma->ma_lmv_size = info->mti_big_lmmsize;
ma->ma_valid = 0;
rc = mdt_stripe_get(info, obj, ma, XATTR_NAME_LMV);
if (rc)
GOTO(unlock_obj, rc);
/* user may run 'lfs migrate' multiple times, so it's shrunk already */
if (!(ma->ma_valid & MA_LMV))
GOTO(unlock_obj, rc = -EALREADY);
lmv = &ma->ma_lmv->lmv_md_v1;
/* ditto */
if (!(le32_to_cpu(lmv->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION))
GOTO(unlock_obj, rc = -EALREADY);
lum_stripe_count = lmu->lum_stripe_count;
if (!lum_stripe_count)
lum_stripe_count = cpu_to_le32(1);
if (lmv->lmv_migrate_offset != lum_stripe_count) {
CERROR("%s: "DFID" migrate mdt count mismatch %u != %u\n",
mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
lmv->lmv_migrate_offset, lmu->lum_stripe_count);
GOTO(unlock_obj, rc = -EINVAL);
}
if (lmv->lmv_master_mdt_index != lmu->lum_stripe_offset) {
CERROR("%s: "DFID" migrate mdt index mismatch %u != %u\n",
mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
lmv->lmv_master_mdt_index, lmu->lum_stripe_offset);
GOTO(unlock_obj, rc = -EINVAL);
}
if (lum_stripe_count > 1 &&
(lmv->lmv_hash_type & cpu_to_le32(LMV_HASH_TYPE_MASK)) !=
lmu->lum_hash_type) {
CERROR("%s: "DFID" migrate mdt hash mismatch %u != %u\n",
mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
lmv->lmv_hash_type, lmu->lum_hash_type);
GOTO(unlock_obj, rc = -EINVAL);
}
buf->lb_buf = rr->rr_eadata;
buf->lb_len = rr->rr_eadatalen;
rc = mo_xattr_set(env, mdt_object_child(obj), buf, XATTR_NAME_LMV, 0);
GOTO(unlock_obj, rc);
unlock_obj:
mdt_reint_striped_unlock(info, obj, lhc, einfo, rc);
unlock_pobj:
mdt_object_unlock(info, pobj, lhp, rc);
put_pobj:
mdt_object_put(env, pobj);
put_obj:
mdt_object_put(env, obj);
return rc;
}
/*
* VBR: save parent version in reply and child version getting by its name.
* Version of child is getting and checking during its lookup. If
*/
static int mdt_reint_unlink(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct lu_fid *child_fid = &info->mti_tmp_fid1;
struct mdt_object *mp;
struct mdt_object *mc;
struct mdt_lock_handle *parent_lh;
struct mdt_lock_handle *child_lh;
struct lu_name *lname;
int rc;
int no_name = 0;
ENTRY;
DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
rr->rr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
RETURN(err_serious(-ENOENT));
if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1))
RETURN(-EPERM);
/*
* step 1: Found the parent.
*/
mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(mp)) {
rc = PTR_ERR(mp);
GOTO(out, rc);
}
parent_lh = &info->mti_lh[MDT_LH_PARENT];
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
if (mdt_object_remote(mp)) {
mdt_lock_reg_init(parent_lh, LCK_EX);
rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
parent_lh->mlh_rreg_mode,
MDS_INODELOCK_UPDATE);
if (rc != ELDLM_OK)
GOTO(put_parent, rc);
} else {
mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
rr->rr_namelen);
rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc)
GOTO(put_parent, rc);
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(unlock_parent, rc);
}
/* step 2: find & lock the child */
/* lookup child object along with version checking */
fid_zero(child_fid);
rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
if (rc != 0) {
/* Name might not be able to find during resend of
* remote unlink, considering following case.
* dir_A is a remote directory, the name entry of
* dir_A is on MDT0, the directory is on MDT1,
*
* 1. client sends unlink req to MDT1.
* 2. MDT1 sends name delete update to MDT0.
* 3. name entry is being deleted in MDT0 synchronously.
* 4. MDT1 is restarted.
* 5. client resends unlink req to MDT1. So it can not
* find the name entry on MDT0 anymore.
* In this case, MDT1 only needs to destory the local
* directory.
* */
if (mdt_object_remote(mp) && rc == -ENOENT &&
!fid_is_zero(rr->rr_fid2) &&
lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
no_name = 1;
*child_fid = *rr->rr_fid2;
} else {
GOTO(unlock_parent, rc);
}
}
if (fid_is_obf(child_fid) || fid_is_dot_lustre(child_fid))
GOTO(unlock_parent, rc = -EPERM);
mdt_reint_init_ma(info, ma);
/* We will lock the child regardless it is local or remote. No harm. */
mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
if (IS_ERR(mc))
GOTO(unlock_parent, rc = PTR_ERR(mc));
child_lh = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(child_lh, LCK_EX);
if (mdt_object_remote(mc)) {
struct mdt_body *repbody;
if (!fid_is_zero(rr->rr_fid2)) {
CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
mdt_obd_name(info->mti_mdt),
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
GOTO(put_child, rc = -ENOENT);
}
CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
mdt_obd_name(info->mti_mdt),
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
if (!mdt_is_dne_client(req->rq_export))
/* Return -EIO for old client */
GOTO(put_child, rc = -EIO);
if (info->mti_spec.sp_rm_entry) {
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: unlink remote entry is only "
"permitted for administrator: rc = %d\n",
mdt_obd_name(info->mti_mdt),
-EPERM);
GOTO(put_child, rc = -EPERM);
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
NULL, lname, ma, no_name);
GOTO(put_child, rc);
}
/* Revoke the LOOKUP lock of the remote object granted by
* this MDT. Since the unlink will happen on another MDT,
* it will release the LOOKUP lock right away. Then What
* would happen if another client try to grab the LOOKUP
* lock at the same time with unlink XXX */
mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
LASSERT(repbody != NULL);
repbody->fid1 = *mdt_object_fid(mc);
repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
GOTO(unlock_child, rc = -EREMOTE);
} else if (info->mti_spec.sp_rm_entry) {
rc = -EPERM;
CDEBUG(D_INFO, "%s: no rm_entry on local dir '%s': rc = %d\n",
mdt_obd_name(info->mti_mdt), (char *)rr->rr_name, rc);
GOTO(put_child, rc);
}
/* We used to acquire MDS_INODELOCK_FULL here but we can't do
* this now because a running HSM restore on the child (unlink
* victim) will hold the layout lock. See LU-4002. */
rc = mdt_object_lock(info, mc, child_lh,
MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
MDT_CROSS_LOCK);
if (rc != 0)
GOTO(put_child, rc);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_UNLINK_WRITE);
/* save version when object is locked */
mdt_version_get_save(info, mc, 1);
/*
* Now we can only make sure we need MA_INODE, in mdd layer, will check
* whether need MA_LOV and MA_COOKIE.
*/
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
mutex_lock(&mc->mot_lov_mutex);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
mdt_object_child(mc), lname, ma, no_name);
mutex_unlock(&mc->mot_lov_mutex);
if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
rc = mdt_attr_get_complex(info, mc, ma);
if (rc == 0)
mdt_handle_last_unlink(info, mc, ma);
if (ma->ma_valid & MA_INODE) {
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
mdt_counter_incr(req, LPROC_MDT_RMDIR);
break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
mdt_counter_incr(req, LPROC_MDT_UNLINK);
break;
default:
LASSERTF(0, "bad file type %o unlinking\n",
ma->ma_attr.la_mode);
}
}
EXIT;
unlock_child:
mdt_object_unlock(info, mc, child_lh, rc);
put_child:
mdt_object_put(info->mti_env, mc);
unlock_parent:
mdt_object_unlock(info, mp, parent_lh, rc);
put_parent:
mdt_object_put(info->mti_env, mp);
out:
return rc;
}
/*
* VBR: save versions in reply: 0 - parent; 1 - child by fid; 2 - target by
* name.
*/
static int mdt_reint_link(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct mdt_object *ms;
struct mdt_object *mp;
struct mdt_lock_handle *lhs;
struct mdt_lock_handle *lhp;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/%s",
PFID(rr->rr_fid1), PFID(rr->rr_fid2), rr->rr_name);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
RETURN(err_serious(-ENOENT));
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
/* Invalid case so return error immediately instead of
* processing it */
if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
RETURN(-EPERM);
if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1) ||
fid_is_obf(rr->rr_fid2) || fid_is_dot_lustre(rr->rr_fid2))
RETURN(-EPERM);
/* step 1: find & lock the target parent dir */
lhp = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lhp, LCK_PW, rr->rr_name,
rr->rr_namelen);
mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
MDS_INODELOCK_UPDATE);
if (IS_ERR(mp))
RETURN(PTR_ERR(mp));
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(out_unlock_parent, rc);
/* step 2: find & lock the source */
lhs = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(lhs, LCK_EX);
ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(ms))
GOTO(out_unlock_parent, rc = PTR_ERR(ms));
if (mdt_object_remote(ms)) {
mdt_object_put(info->mti_env, ms);
CERROR("%s: source inode "DFID" on remote MDT from "DFID"\n",
mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
PFID(rr->rr_fid2));
GOTO(out_unlock_parent, rc = -EXDEV);
}
rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE |
MDS_INODELOCK_XATTR, MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, ms);
GOTO(out_unlock_parent, rc);
}
/* step 3: link it */
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_LINK_WRITE);
tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(ms));
rc = mdt_version_get_check_save(info, ms, 1);
if (rc)
GOTO(out_unlock_child, rc);
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
/** check target version by name during replay */
rc = mdt_lookup_version_check(info, mp, lname, &info->mti_tmp_fid1, 2);
if (rc != 0 && rc != -ENOENT)
GOTO(out_unlock_child, rc);
/* save version of file name for replay, it must be ENOENT here */
if (!req_is_replay(mdt_info_req(info))) {
if (rc != -ENOENT) {
CDEBUG(D_INFO, "link target %.*s existed!\n",
rr->rr_namelen, (char *)rr->rr_name);
GOTO(out_unlock_child, rc = -EEXIST);
}
info->mti_ver[2] = ENOENT_VERSION;
mdt_version_save(mdt_info_req(info), info->mti_ver[2], 2);
}
rc = mdo_link(info->mti_env, mdt_object_child(mp),
mdt_object_child(ms), lname, ma);
if (rc == 0)
mdt_counter_incr(req, LPROC_MDT_LINK);
EXIT;
out_unlock_child:
mdt_object_unlock_put(info, ms, lhs, rc);
out_unlock_parent:
mdt_object_unlock_put(info, mp, lhp, rc);
return rc;
}
/**
* Object updates between Targets. Because all the updates has been
* dis-assemblied into object updates in master MDD layer, so out
* will skip MDD layer, and call OSD API directly to execute these
* updates.
*
* In 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 mdt_thread_info *info)
{
struct req_capsule *pill = info->mti_pill;
struct update_buf *ubuf;
struct update *update;
struct thandle_exec_args *th = &info->mti_handle;
int bufsize;
int count;
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",
mdt_obd_name(info->mti_mdt), bufsize, -EPROTO);
RETURN(err_serious(-EPROTO));
}
ubuf = req_capsule_client_get(pill, &RMF_UPDATE);
if (ubuf == NULL) {
CERROR("%s: No buf!: rc = %d\n", mdt_obd_name(info->mti_mdt),
-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",
mdt_obd_name(info->mti_mdt), 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",
mdt_obd_name(info->mti_mdt), -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",
mdt_obd_name(info->mti_mdt), rc);
RETURN(rc);
}
/* Prepare the update reply buffer */
info->mti_u.update.mti_update_reply =
req_capsule_server_get(pill, &RMF_UPDATE_REPLY);
update_init_reply_buf(info->mti_u.update.mti_update_reply, count);
rc = out_tx_start(info->mti_env, info->mti_mdt, th);
if (rc != 0)
RETURN(rc);
/* 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 out_handler *h;
struct dt_object *dt_obj;
update = (struct update *)((char *)ubuf + off);
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",
mdt_obd_name(info->mti_mdt),
PFID(&update->u_fid), -EPROTO);
GOTO(out, rc = err_serious(-EPROTO));
}
dt_obj = out_object_find(info, &update->u_fid);
if (IS_ERR(dt_obj))
GOTO(out, rc = PTR_ERR(dt_obj));
info->mti_u.update.mti_dt_object = dt_obj;
info->mti_u.update.mti_update = update;
info->mti_u.update.mti_update_reply_index = i;
h = mdt_handler_find(update->u_type, out_handlers);
if (likely(h != NULL)) {
rc = h->mh_act(info);
} else {
CERROR("%s: The unsupported opc: 0x%x\n",
mdt_obd_name(info->mti_mdt), update->u_type);
lu_object_put(info->mti_env, &dt_obj->do_lu);
GOTO(out, rc = -ENOTSUPP);
}
lu_object_put(info->mti_env, &dt_obj->do_lu);
if (rc < 0)
GOTO(out, rc);
off += cfs_size_round(update_size(update));
}
out:
rc1 = out_tx_end(info, th);
rc = rc == 0 ? rc1 : rc;
info->mti_fail_id = OBD_FAIL_UPDATE_OBJ_NET;
RETURN(rc);
}
/**
* Process the HSM actions described in a struct hsm_user_request.
*
* The action described in hur will be send to coordinator to be saved and
* processed later or either handled directly if hur.hur_action is HUA_RELEASE.
*
* This is MDS_HSM_REQUEST RPC handler.
*/
int mdt_hsm_request(struct tgt_session_info *tsi)
{
struct mdt_thread_info *info;
struct req_capsule *pill = tsi->tsi_pill;
struct hsm_request *hr;
struct hsm_user_item *hui;
struct hsm_action_list *hal;
struct hsm_action_item *hai;
const void *data;
int hui_list_size;
int data_size;
enum hsm_copytool_action action = HSMA_NONE;
__u64 compound_id;
int hal_size, i, rc;
ENTRY;
hr = req_capsule_client_get(pill, &RMF_MDS_HSM_REQUEST);
hui = req_capsule_client_get(pill, &RMF_MDS_HSM_USER_ITEM);
data = req_capsule_client_get(pill, &RMF_GENERIC_DATA);
if (tsi->tsi_mdt_body == NULL || hr == NULL || hui == NULL || data == NULL)
RETURN(-EPROTO);
/* Sanity check. Nothing to do with an empty list */
if (hr->hr_itemcount == 0)
RETURN(0);
hui_list_size = req_capsule_get_size(pill, &RMF_MDS_HSM_USER_ITEM,
RCL_CLIENT);
if (hui_list_size < hr->hr_itemcount * sizeof(*hui))
RETURN(-EPROTO);
data_size = req_capsule_get_size(pill, &RMF_GENERIC_DATA, RCL_CLIENT);
if (data_size != hr->hr_data_len)
RETURN(-EPROTO);
info = tsi2mdt_info(tsi);
/* Only valid if client is remote */
rc = mdt_init_ucred(info, (struct mdt_body *)info->mti_body);
if (rc)
GOTO(out, rc);
switch (hr->hr_action) {
/* code to be removed in hsm1_merge and final patch */
case HUA_RELEASE:
CERROR("Release action is not working in hsm1_coord\n");
GOTO(out_ucred, rc = -EINVAL);
break;
/* end of code to be removed */
case HUA_ARCHIVE:
action = HSMA_ARCHIVE;
break;
case HUA_RESTORE:
action = HSMA_RESTORE;
break;
case HUA_REMOVE:
action = HSMA_REMOVE;
break;
case HUA_CANCEL:
action = HSMA_CANCEL;
break;
default:
CERROR("Unknown hsm action: %d\n", hr->hr_action);
GOTO(out_ucred, rc = -EINVAL);
}
hal_size = sizeof(*hal) + cfs_size_round(MTI_NAME_MAXLEN) /* fsname */ +
(sizeof(*hai) + cfs_size_round(hr->hr_data_len)) *
hr->hr_itemcount;
MDT_HSM_ALLOC(hal, hal_size);
if (hal == NULL)
GOTO(out_ucred, rc = -ENOMEM);
hal->hal_version = HAL_VERSION;
hal->hal_archive_id = hr->hr_archive_id;
hal->hal_flags = hr->hr_flags;
obd_uuid2fsname(hal->hal_fsname, mdt_obd_name(info->mti_mdt),
MTI_NAME_MAXLEN);
hal->hal_count = 0;
hai = hai_first(hal);
for (i = 0; i < hr->hr_itemcount; i++, hai = hai_next(hai)) {
/* Get rid of duplicate entries. Otherwise we get
* duplicated work in the llog. */
if (is_fid_in_hal(hal, &hui[i].hui_fid))
continue;
hai->hai_action = action;
hai->hai_cookie = 0;
hai->hai_gid = 0;
hai->hai_fid = hui[i].hui_fid;
hai->hai_extent = hui[i].hui_extent;
memcpy(hai->hai_data, data, hr->hr_data_len);
hai->hai_len = sizeof(*hai) + hr->hr_data_len;
hal->hal_count++;
}
rc = mdt_hsm_add_actions(info, hal, &compound_id);
MDT_HSM_FREE(hal, hal_size);
GOTO(out_ucred, rc);
out_ucred:
mdt_exit_ucred(info);
out:
mdt_thread_info_fini(info);
return rc;
}
int mdt_reint_setxattr(struct mdt_thread_info *info,
struct mdt_lock_handle *unused)
{
struct ptlrpc_request *req = mdt_info_req(info);
struct lu_ucred *uc = lu_ucred(info->mti_env);
struct mdt_lock_handle *lh;
const struct lu_env *env = info->mti_env;
struct lu_buf *buf = &info->mti_buf;
struct mdt_reint_record *rr = &info->mti_rr;
struct md_attr *ma = &info->mti_attr;
struct lu_attr *attr = &info->mti_attr.ma_attr;
struct mdt_object *obj;
struct md_object *child;
__u64 valid = attr->la_valid;
const char *xattr_name = rr->rr_name.ln_name;
int xattr_len = rr->rr_eadatalen;
__u64 lockpart;
int rc;
posix_acl_xattr_header *new_xattr = NULL;
__u32 remote = exp_connect_rmtclient(info->mti_exp);
__u32 perm;
ENTRY;
CDEBUG(D_INODE, "setxattr for "DFID"\n", PFID(rr->rr_fid1));
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SETXATTR))
RETURN(err_serious(-ENOMEM));
CDEBUG(D_INODE, "%s xattr %s\n",
valid & OBD_MD_FLXATTR ? "set" : "remove", xattr_name);
rc = mdt_init_ucred_reint(info);
if (rc != 0)
RETURN(rc);
if (valid & OBD_MD_FLRMTRSETFACL) {
if (unlikely(!remote))
GOTO(out, rc = err_serious(-EINVAL));
perm = mdt_identity_get_perm(uc->uc_identity, remote,
req->rq_peer.nid);
if (!(perm & CFS_RMTACL_PERM))
GOTO(out, rc = err_serious(-EPERM));
}
if (strncmp(xattr_name, XATTR_USER_PREFIX,
sizeof(XATTR_USER_PREFIX) - 1) == 0) {
if (!(exp_connect_flags(req->rq_export) & OBD_CONNECT_XATTR))
GOTO(out, rc = -EOPNOTSUPP);
} else if (strncmp(xattr_name, XATTR_TRUSTED_PREFIX,
sizeof(XATTR_TRUSTED_PREFIX) - 1) == 0) {
if (!md_capable(mdt_ucred(info), CFS_CAP_SYS_ADMIN))
GOTO(out, rc = -EPERM);
if (strcmp(xattr_name, XATTR_NAME_LOV) == 0 ||
strcmp(xattr_name, XATTR_NAME_LMA) == 0 ||
strcmp(xattr_name, XATTR_NAME_LMV) == 0 ||
strcmp(xattr_name, XATTR_NAME_LINK) == 0 ||
strcmp(xattr_name, XATTR_NAME_FID) == 0 ||
strcmp(xattr_name, XATTR_NAME_VERSION) == 0 ||
strcmp(xattr_name, XATTR_NAME_SOM) == 0 ||
strcmp(xattr_name, XATTR_NAME_HSM) == 0 ||
strcmp(xattr_name, XATTR_NAME_LFSCK_NAMESPACE) == 0)
GOTO(out, rc = 0);
} else if ((valid & OBD_MD_FLXATTR) &&
(strcmp(xattr_name, XATTR_NAME_ACL_ACCESS) == 0 ||
strcmp(xattr_name, XATTR_NAME_ACL_DEFAULT) == 0)) {
/* currently lustre limit acl access size */
if (xattr_len > LUSTRE_POSIX_ACL_MAX_SIZE)
GOTO(out, rc = -ERANGE);
}
lockpart = MDS_INODELOCK_UPDATE;
/* Revoke all clients' lookup lock, since the access
* permissions for this inode is changed when ACL_ACCESS is
* set. This isn't needed for ACL_DEFAULT, since that does
* not change the access permissions of this inode, nor any
* other existing inodes. It is setting the ACLs inherited
* by new directories/files at create time. */
/* We need revoke both LOOKUP|PERM lock here, see mdt_attr_set. */
if (!strcmp(xattr_name, XATTR_NAME_ACL_ACCESS))
lockpart |= MDS_INODELOCK_PERM | MDS_INODELOCK_LOOKUP;
/* We need to take the lock on behalf of old clients so that newer
* clients flush their xattr caches */
else
lockpart |= MDS_INODELOCK_XATTR;
lh = &info->mti_lh[MDT_LH_PARENT];
/* ACLs were sent to clients under LCK_CR locks, so taking LCK_EX
* to cancel them. */
mdt_lock_reg_init(lh, LCK_EX);
obj = mdt_object_find_lock(info, rr->rr_fid1, lh, lockpart);
if (IS_ERR(obj))
GOTO(out, rc = PTR_ERR(obj));
tgt_vbr_obj_set(env, mdt_obj2dt(obj));
rc = mdt_version_get_check_save(info, obj, 0);
if (rc)
GOTO(out_unlock, rc);
if (unlikely(!(valid & OBD_MD_FLCTIME))) {
/* This isn't strictly an error, but all current clients
* should set OBD_MD_FLCTIME when setting attributes. */
CWARN("%s: client miss to set OBD_MD_FLCTIME when "
"setxattr %s: [object "DFID"] [valid "LPU64"]\n",
mdt_obd_name(info->mti_mdt), xattr_name,
PFID(rr->rr_fid1), valid);
attr->la_ctime = cfs_time_current_sec();
}
attr->la_valid = LA_CTIME;
child = mdt_object_child(obj);
if (valid & OBD_MD_FLXATTR) {
char *xattr = (void *)rr->rr_eadata;
if (xattr_len > 0) {
int flags = 0;
if (valid & OBD_MD_FLRMTLSETFACL) {
if (unlikely(!remote))
GOTO(out_unlock, rc = -EINVAL);
xattr_len = mdt_rmtlsetfacl(info, child,
xattr_name,
(ext_acl_xattr_header *)xattr,
&new_xattr);
if (xattr_len < 0)
GOTO(out_unlock, rc = xattr_len);
xattr = (char *)new_xattr;
}
if (attr->la_flags & XATTR_REPLACE)
flags |= LU_XATTR_REPLACE;
if (attr->la_flags & XATTR_CREATE)
flags |= LU_XATTR_CREATE;
mdt_fail_write(env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_SETXATTR_WRITE);
buf->lb_buf = xattr;
buf->lb_len = xattr_len;
rc = mo_xattr_set(env, child, buf, xattr_name, flags);
/* update ctime after xattr changed */
if (rc == 0) {
ma->ma_attr_flags |= MDS_PERM_BYPASS;
mo_attr_set(env, child, ma);
}
}
} else if (valid & OBD_MD_FLXATTRRM) {
rc = mo_xattr_del(env, child, xattr_name);
/* update ctime after xattr changed */
if (rc == 0) {
ma->ma_attr_flags |= MDS_PERM_BYPASS;
mo_attr_set(env, child, ma);
}
} else {
CDEBUG(D_INFO, "valid bits: "LPX64"\n", valid);
rc = -EINVAL;
}
if (rc == 0)
mdt_counter_incr(req, LPROC_MDT_SETXATTR);
EXIT;
out_unlock:
mdt_object_unlock_put(info, obj, lh, rc);
if (unlikely(new_xattr != NULL))
lustre_posix_acl_xattr_free(new_xattr, xattr_len);
out:
mdt_exit_ucred(info);
return rc;
}
/*
* VBR: we save three versions in reply:
* 0 - parent. Check that parent version is the same during replay.
* 1 - name. Version of 'name' if file exists with the same name or
* ENOENT_VERSION, it is needed because file may appear due to missed replays.
* 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
* check.
*/
static int mdt_md_create(struct mdt_thread_info *info)
{
struct mdt_device *mdt = info->mti_mdt;
struct mdt_object *parent;
struct mdt_object *child;
struct mdt_lock_handle *lh;
struct mdt_body *repbody;
struct md_attr *ma = &info->mti_attr;
struct mdt_reint_record *rr = &info->mti_rr;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1))
RETURN(-EPERM);
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
lh = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
MDS_INODELOCK_UPDATE);
if (IS_ERR(parent))
RETURN(PTR_ERR(parent));
rc = mdt_version_get_check_save(info, parent, 0);
if (rc)
GOTO(out_put_parent, rc);
/*
* Check child name version during replay.
* During create replay a file may exist with same name.
*/
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
rc = mdt_lookup_version_check(info, parent, lname,
&info->mti_tmp_fid1, 1);
if (rc == 0)
GOTO(out_put_parent, rc = -EEXIST);
/* -ENOENT is expected here */
if (rc != -ENOENT)
GOTO(out_put_parent, rc);
/* save version of file name for replay, it must be ENOENT here */
mdt_enoent_version_save(info, 1);
child = mdt_object_new(info->mti_env, mdt, rr->rr_fid2);
if (likely(!IS_ERR(child))) {
struct md_object *next = mdt_object_child(parent);
if (mdt_object_remote(child)) {
struct seq_server_site *ss;
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
if (uc->uc_gid !=
mdt->mdt_enable_remote_dir_gid &&
mdt->mdt_enable_remote_dir_gid != -1) {
CERROR("%s: Creating remote dir is only"
" permitted for administrator or"
" set mdt_enable_remote_dir_gid:"
" rc = %d\n",
mdt_obd_name(mdt), -EPERM);
GOTO(out_put_child, rc = -EPERM);
}
}
ss = mdt_seq_site(mdt);
if (ss->ss_node_id != 0 &&
mdt->mdt_enable_remote_dir == 0) {
CERROR("%s: remote dir is only permitted on"
" MDT0 or set_param"
" mdt.*.enable_remote_dir=1\n",
mdt_obd_name(mdt));
GOTO(out_put_child, rc = -EPERM);
}
if (!mdt_is_dne_client(mdt_info_req(info)->rq_export)) {
/* Return -EIO for old client */
GOTO(out_put_child, rc = -EIO);
}
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
/* capa for cross-ref will be stored here */
ma->ma_capa = req_capsule_server_get(info->mti_pill,
&RMF_CAPA1);
LASSERT(ma->ma_capa);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_CREATE_WRITE);
/* Version of child will be updated on disk. */
tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
rc = mdt_version_get_check_save(info, child, 2);
if (rc)
GOTO(out_put_child, rc);
/* Let lower layer know current lock mode. */
info->mti_spec.sp_cr_mode =
mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
/*
* Do not perform lookup sanity check. We know that name does
* not exist.
*/
info->mti_spec.sp_cr_lookup = 0;
info->mti_spec.sp_feat = &dt_directory_features;
rc = mdo_create(info->mti_env, next, lname,
mdt_object_child(child),
&info->mti_spec, ma);
if (rc == 0)
rc = mdt_attr_get_complex(info, child, ma);
if (rc == 0) {
/* Return fid & attr to client. */
if (ma->ma_valid & MA_INODE)
mdt_pack_attr2body(info, repbody, &ma->ma_attr,
mdt_object_fid(child));
}
out_put_child:
mdt_object_put(info->mti_env, child);
} else {
rc = PTR_ERR(child);
}
mdt_create_pack_capa(info, rc, child, repbody);
out_put_parent:
mdt_object_unlock_put(info, parent, lh, rc);
RETURN(rc);
}
