static int
zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
dmu_buf_t **db, void *tag)
{
dmu_object_info_t doi;
int error;
if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
return (error);
dmu_object_info_from_db(*db, &doi);
if ((doi.doi_bonus_type != DMU_OT_SA &&
doi.doi_bonus_type != DMU_OT_ZNODE) ||
(doi.doi_bonus_type == DMU_OT_ZNODE &&
doi.doi_bonus_size < sizeof (znode_phys_t))) {
sa_buf_rele(*db, tag);
return (SET_ERROR(ENOTSUP));
}
error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
if (error != 0) {
sa_buf_rele(*db, tag);
return (error);
}
return (0);
}
int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
uint64_t oid, dmu_buf_t **dbp, void *tag)
{
dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
int rc;
LASSERT(tag);
rc = -sa_buf_hold(os, oid, tag, dbp);
if (rc)
return rc;
dmu_object_info_from_db(*dbp, doi);
if (unlikely (oid != DMU_USERUSED_OBJECT &&
oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
sa_buf_rele(*dbp, tag);
*dbp = NULL;
return -EINVAL;
}
LASSERT(*dbp);
LASSERT((*dbp)->db_object == oid);
LASSERT((*dbp)->db_offset == -1);
LASSERT((*dbp)->db_data != NULL);
return 0;
}
static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
{
struct osd_object *obj = osd_obj(l);
if (obj->oo_db != NULL) {
osd_object_sa_fini(obj);
if (obj->oo_sa_xattr) {
nvlist_free(obj->oo_sa_xattr);
obj->oo_sa_xattr = NULL;
}
sa_buf_rele(obj->oo_db, osd_obj_tag);
cfs_list_del(&obj->oo_sa_linkage);
obj->oo_db = NULL;
}
}
/*
* The transaction passed to this routine must have
* dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
* to a transaction group.
*/
int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
uint64_t parent)
{
uint64_t oid;
int rc;
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
/* Assert that the transaction has been assigned to a
transaction group. */
LASSERT(tx->tx_txg != 0);
/* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
* would get an additional ditto copy */
if (unlikely(S_ISREG(la->la_mode) &&
fid_seq_is_local_file(fid_seq(fid))))
type = DMU_OTN_UINT8_METADATA;
/* Create a new DMU object. */
oid = dmu_object_alloc(osd->od_os, type, 0,
DMU_OT_SA, DN_MAX_BONUSLEN, tx);
rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
LASSERT(la->la_valid & LA_MODE);
la->la_size = 0;
la->la_nlink = 1;
rc = __osd_attr_init(env, osd, oid, tx, la, parent);
if (rc != 0) {
sa_buf_rele(*dbp, osd_obj_tag);
*dbp = NULL;
dmu_object_free(osd->od_os, oid, tx);
return rc;
}
return 0;
}
int
zfs_zget(zfs_sb_t *zsb, uint64_t obj_num, znode_t **zpp)
{
dmu_object_info_t doi;
dmu_buf_t *db;
znode_t *zp;
int err;
sa_handle_t *hdl;
*zpp = NULL;
again:
ZFS_OBJ_HOLD_ENTER(zsb, obj_num);
err = sa_buf_hold(zsb->z_os, obj_num, NULL, &db);
if (err) {
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (err);
}
dmu_object_info_from_db(db, &doi);
if (doi.doi_bonus_type != DMU_OT_SA &&
(doi.doi_bonus_type != DMU_OT_ZNODE ||
(doi.doi_bonus_type == DMU_OT_ZNODE &&
doi.doi_bonus_size < sizeof (znode_phys_t)))) {
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (SET_ERROR(EINVAL));
}
hdl = dmu_buf_get_user(db);
if (hdl != NULL) {
zp = sa_get_userdata(hdl);
/*
* Since "SA" does immediate eviction we
* should never find a sa handle that doesn't
* know about the znode.
*/
ASSERT3P(zp, !=, NULL);
mutex_enter(&zp->z_lock);
ASSERT3U(zp->z_id, ==, obj_num);
if (zp->z_unlinked) {
err = SET_ERROR(ENOENT);
} else {
/*
* If igrab() returns NULL the VFS has independently
* determined the inode should be evicted and has
* called iput_final() to start the eviction process.
* The SA handle is still valid but because the VFS
* requires that the eviction succeed we must drop
* our locks and references to allow the eviction to
* complete. The zfs_zget() may then be retried.
*
* This unlikely case could be optimized by registering
* a sops->drop_inode() callback. The callback would
* need to detect the active SA hold thereby informing
* the VFS that this inode should not be evicted.
*/
if (igrab(ZTOI(zp)) == NULL) {
mutex_exit(&zp->z_lock);
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
goto again;
}
*zpp = zp;
err = 0;
}
mutex_exit(&zp->z_lock);
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (err);
}
int
zfs_rezget(znode_t *zp)
{
zfs_sb_t *zsb = ZTOZSB(zp);
dmu_object_info_t doi;
dmu_buf_t *db;
uint64_t obj_num = zp->z_id;
uint64_t mode;
sa_bulk_attr_t bulk[8];
int err;
int count = 0;
uint64_t gen;
ZFS_OBJ_HOLD_ENTER(zsb, obj_num);
mutex_enter(&zp->z_acl_lock);
if (zp->z_acl_cached) {
zfs_acl_free(zp->z_acl_cached);
zp->z_acl_cached = NULL;
}
mutex_exit(&zp->z_acl_lock);
rw_enter(&zp->z_xattr_lock, RW_WRITER);
if (zp->z_xattr_cached) {
nvlist_free(zp->z_xattr_cached);
zp->z_xattr_cached = NULL;
}
if (zp->z_xattr_parent) {
iput(ZTOI(zp->z_xattr_parent));
zp->z_xattr_parent = NULL;
}
rw_exit(&zp->z_xattr_lock);
ASSERT(zp->z_sa_hdl == NULL);
err = sa_buf_hold(zsb->z_os, obj_num, NULL, &db);
if (err) {
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (err);
}
dmu_object_info_from_db(db, &doi);
if (doi.doi_bonus_type != DMU_OT_SA &&
(doi.doi_bonus_type != DMU_OT_ZNODE ||
(doi.doi_bonus_type == DMU_OT_ZNODE &&
doi.doi_bonus_size < sizeof (znode_phys_t)))) {
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (SET_ERROR(EINVAL));
}
zfs_znode_sa_init(zsb, zp, db, doi.doi_bonus_type, NULL);
/* reload cached values */
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zsb), NULL,
&gen, sizeof (gen));
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zsb), NULL,
&zp->z_size, sizeof (zp->z_size));
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zsb), NULL,
&zp->z_links, sizeof (zp->z_links));
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zsb), NULL,
&zp->z_pflags, sizeof (zp->z_pflags));
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zsb), NULL,
&zp->z_atime, sizeof (zp->z_atime));
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL,
&zp->z_uid, sizeof (zp->z_uid));
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb), NULL,
&zp->z_gid, sizeof (zp->z_gid));
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL,
&mode, sizeof (mode));
if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
zfs_znode_dmu_fini(zp);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (SET_ERROR(EIO));
}
zp->z_mode = mode;
if (gen != zp->z_gen) {
zfs_znode_dmu_fini(zp);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (SET_ERROR(EIO));
}
zp->z_unlinked = (zp->z_links == 0);
zp->z_blksz = doi.doi_data_block_size;
zfs_inode_update(zp);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (0);
}
int
zfs_zget(zfs_sb_t *zsb, uint64_t obj_num, znode_t **zpp)
{
dmu_object_info_t doi;
dmu_buf_t *db;
znode_t *zp;
int err;
sa_handle_t *hdl;
struct inode *ip;
*zpp = NULL;
again:
ip = ilookup(zsb->z_sb, obj_num);
ZFS_OBJ_HOLD_ENTER(zsb, obj_num);
err = sa_buf_hold(zsb->z_os, obj_num, NULL, &db);
if (err) {
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
iput(ip);
return (err);
}
dmu_object_info_from_db(db, &doi);
if (doi.doi_bonus_type != DMU_OT_SA &&
(doi.doi_bonus_type != DMU_OT_ZNODE ||
(doi.doi_bonus_type == DMU_OT_ZNODE &&
doi.doi_bonus_size < sizeof (znode_phys_t)))) {
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
iput(ip);
return (SET_ERROR(EINVAL));
}
hdl = dmu_buf_get_user(db);
if (hdl != NULL) {
if (ip == NULL) {
/*
* ilookup returned NULL, which means
* the znode is dying - but the SA handle isn't
* quite dead yet, we need to drop any locks
* we're holding, re-schedule the task and try again.
*/
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
schedule();
goto again;
}
zp = sa_get_userdata(hdl);
/*
* Since "SA" does immediate eviction we
* should never find a sa handle that doesn't
* know about the znode.
*/
ASSERT3P(zp, !=, NULL);
mutex_enter(&zp->z_lock);
ASSERT3U(zp->z_id, ==, obj_num);
if (zp->z_unlinked) {
err = SET_ERROR(ENOENT);
} else {
igrab(ZTOI(zp));
*zpp = zp;
err = 0;
}
sa_buf_rele(db, NULL);
mutex_exit(&zp->z_lock);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
iput(ip);
return (err);
}
ASSERT3P(ip, ==, NULL);
/*
* Not found create new znode/vnode but only if file exists.
*
* There is a small window where zfs_vget() could
* find this object while a file create is still in
* progress. This is checked for in zfs_znode_alloc()
*
* if zfs_znode_alloc() fails it will drop the hold on the
* bonus buffer.
*/
zp = zfs_znode_alloc(zsb, db, doi.doi_data_block_size,
doi.doi_bonus_type, obj_num, NULL, NULL);
if (zp == NULL) {
err = SET_ERROR(ENOENT);
} else {
*zpp = zp;
}
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (err);
}
void
zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
{
sa_handle_destroy(hdl);
sa_buf_rele(db, tag);
}
/*
* Concurrency: @dt is write locked.
*/
static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
struct lu_attr *attr,
struct dt_allocation_hint *hint,
struct dt_object_format *dof,
struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
char *buf = info->oti_str;
struct osd_thandle *oh;
dmu_buf_t *db = NULL;
uint64_t zapid, parent = 0;
int rc;
ENTRY;
/* concurrent create declarations should not see
* the object inconsistent (db, attr, etc).
* in regular cases acquisition should be cheap */
down_write(&obj->oo_guard);
if (unlikely(dt_object_exists(dt)))
GOTO(out, rc = -EEXIST);
LASSERT(osd_invariant(obj));
LASSERT(dof != NULL);
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
/*
* XXX missing: Quote handling.
*/
LASSERT(obj->oo_db == NULL);
/* to follow ZFS on-disk format we need
* to initialize parent dnode properly */
if (hint != NULL && hint->dah_parent != NULL &&
!dt_object_remote(hint->dah_parent))
parent = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
/* we may fix some attributes, better do not change the source */
obj->oo_attr = *attr;
obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
db = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
if (IS_ERR(db)) {
rc = PTR_ERR(db);
db = NULL;
GOTO(out, rc);
}
zde->zde_pad = 0;
zde->zde_dnode = db->db_object;
zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
if (rc)
GOTO(out, rc);
/* Now add in all of the "SA" attributes */
rc = -sa_handle_get(osd->od_os, db->db_object, NULL,
SA_HDL_PRIVATE, &obj->oo_sa_hdl);
if (rc)
GOTO(out, rc);
/* configure new osd object */
obj->oo_db = db;
parent = parent != 0 ? parent : zapid;
rc = __osd_attr_init(env, osd, obj->oo_sa_hdl, oh->ot_tx,
&obj->oo_attr, parent);
if (rc)
GOTO(out, rc);
/* XXX: oo_lma_flags */
obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
smp_mb();
obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
/* no body operations for accounting objects */
obj->oo_dt.do_body_ops = &osd_body_ops;
rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
if (rc)
GOTO(out, rc);
/* initialize LMA */
lustre_lma_init(lma, lu_object_fid(&obj->oo_dt.do_lu), 0, 0);
lustre_lma_swab(lma);
rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
(uchar_t *)lma, sizeof(*lma));
if (rc)
GOTO(out, rc);
rc = __osd_sa_xattr_update(env, obj, oh);
if (rc)
GOTO(out, rc);
/* Add new object to inode accounting.
* Errors are not considered as fatal */
rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
(attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
oh->ot_tx);
if (rc)
CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
"(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
(attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
oh->ot_tx);
if (rc)
CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
"(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
out:
if (unlikely(rc && db)) {
dmu_object_free(osd->od_os, db->db_object, oh->ot_tx);
sa_buf_rele(db, osd_obj_tag);
obj->oo_db = NULL;
}
up_write(&obj->oo_guard);
RETURN(rc);
}