int __osd_xattr_del(const struct lu_env *env, struct osd_object *obj,
const char *name, struct osd_thandle *oh)
{
struct osd_device *osd = osd_obj2dev(obj);
uint64_t xa_data_obj;
int rc;
/* try remove xattr from SA at first */
rc = __osd_sa_xattr_del(env, obj, name, oh);
if (rc != -ENOENT)
return rc;
if (obj->oo_xattr == ZFS_NO_OBJECT)
return 0;
rc = -zap_lookup(osd->od_os, obj->oo_xattr, name, sizeof(uint64_t), 1,
&xa_data_obj);
if (rc == -ENOENT) {
rc = 0;
} else if (rc == 0) {
/*
* Entry exists.
* We'll delete the existing object and ZAP entry.
*/
rc = -dmu_object_free(osd->od_os, xa_data_obj, oh->ot_tx);
if (rc)
return rc;
rc = -zap_remove(osd->od_os, obj->oo_xattr, name, oh->ot_tx);
}
return rc;
}
static int
zpios_dmu_object_free(run_args_t *run_args, objset_t *os, uint64_t obj)
{
struct dmu_tx *tx;
int rc;
tx = dmu_tx_create(os);
dmu_tx_hold_free(tx, obj, 0, DMU_OBJECT_END);
rc = dmu_tx_assign(tx, TXG_WAIT);
if (rc) {
zpios_print(run_args->file,
"dmu_tx_assign() failed: %d\n", rc);
dmu_tx_abort(tx);
return (rc);
}
rc = dmu_object_free(os, obj, tx);
if (rc) {
zpios_print(run_args->file,
"dmu_object_free() failed: %d\n", rc);
dmu_tx_abort(tx);
return (rc);
}
dmu_tx_commit(tx);
return (0);
}
void
dsl_dir_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
{
dsl_dir_t *dd = arg1;
objset_t *mos = dd->dd_pool->dp_meta_objset;
uint64_t val, obj;
ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
/* Remove our reservation. */
val = 0;
dsl_dir_set_reservation_sync(dd, &val, cr, tx);
ASSERT3U(dd->dd_used_bytes, ==, 0);
ASSERT3U(dd->dd_phys->dd_reserved, ==, 0);
VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
VERIFY(0 == zap_remove(mos,
dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
obj = dd->dd_object;
dsl_dir_close(dd, tag);
VERIFY(0 == dmu_object_free(mos, obj, tx));
}
void
dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
{
dsl_dir_t *dd = arg1;
objset_t *mos = dd->dd_pool->dp_meta_objset;
uint64_t obj;
dd_used_t t;
ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
/*
* Remove our reservation. The impl() routine avoids setting the
* actual property, which would require the (already destroyed) ds.
*/
dsl_dir_set_reservation_sync_impl(dd, 0, tx);
ASSERT0(dd->dd_phys->dd_used_bytes);
ASSERT0(dd->dd_phys->dd_reserved);
for (t = 0; t < DD_USED_NUM; t++)
ASSERT0(dd->dd_phys->dd_used_breakdown[t]);
VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
VERIFY(0 == zap_remove(mos,
dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
obj = dd->dd_object;
dsl_dir_close(dd, tag);
VERIFY(0 == dmu_object_free(mos, obj, tx));
}
int osd_xattrs_destroy(const struct lu_env *env,
struct osd_object *obj, struct osd_thandle *oh)
{
struct osd_device *osd = osd_obj2dev(obj);
dmu_tx_t *tx = oh->ot_tx;
zap_attribute_t *za = &osd_oti_get(env)->oti_za;
zap_cursor_t *zc;
uint64_t xid;
int rc;
/* The transaction must have been assigned to a transaction group. */
LASSERT(tx->tx_txg != 0);
if (obj->oo_xattr == ZFS_NO_OBJECT)
return 0; /* Nothing to do for SA xattrs */
/* Free the ZAP holding the xattrs */
rc = osd_zap_cursor_init(&zc, osd->od_os, obj->oo_xattr, 0);
if (rc)
return rc;
while (zap_cursor_retrieve(zc, za) == 0) {
LASSERT(za->za_num_integers == 1);
LASSERT(za->za_integer_length == sizeof(uint64_t));
rc = -zap_lookup(osd->od_os, obj->oo_xattr, za->za_name,
sizeof(uint64_t), 1, &xid);
if (rc) {
CERROR("%s: lookup xattr %s failed: rc = %d\n",
osd->od_svname, za->za_name, rc);
} else {
rc = -dmu_object_free(osd->od_os, xid, tx);
if (rc)
CERROR("%s: free xattr %s failed: rc = %d\n",
osd->od_svname, za->za_name, rc);
}
zap_cursor_advance(zc);
}
osd_zap_cursor_fini(zc);
rc = -dmu_object_free(osd->od_os, obj->oo_xattr, tx);
if (rc)
CERROR("%s: free xattr "LPU64" failed: rc = %d\n",
osd->od_svname, obj->oo_xattr, rc);
return rc;
}
void
zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
{
zfs_sb_t *zsb = ZTOZSB(zp);
objset_t *os = zsb->z_os;
uint64_t obj = zp->z_id;
uint64_t acl_obj = zfs_external_acl(zp);
ZFS_OBJ_HOLD_ENTER(zsb, obj);
if (acl_obj) {
VERIFY(!zp->z_is_sa);
VERIFY(0 == dmu_object_free(os, acl_obj, tx));
}
VERIFY(0 == dmu_object_free(os, obj, tx));
zfs_znode_dmu_fini(zp);
ZFS_OBJ_HOLD_EXIT(zsb, obj);
}
int __osd_object_free(udmu_objset_t *uos, uint64_t oid, dmu_tx_t *tx)
{
LASSERT(uos->objects != 0);
spin_lock(&uos->lock);
uos->objects--;
spin_unlock(&uos->lock);
return -dmu_object_free(uos->os, oid, tx);
}
/*
* 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;
}
void
dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
{
dsl_dataset_t *ds = arg1;
dsl_dir_t *dd = ds->ds_dir;
objset_t *mos = dd->dd_pool->dp_meta_objset;
dsl_prop_setarg_t psa;
uint64_t value = 0;
uint64_t obj;
dd_used_t t;
ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
/* Remove our reservation. */
dsl_prop_setarg_init_uint64(&psa, "reservation",
(ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
&value);
psa.psa_effective_value = 0; /* predict default value */
dsl_dir_set_reservation_sync(ds, &psa, tx);
ASSERT3U(dd->dd_phys->dd_used_bytes, ==, 0);
ASSERT3U(dd->dd_phys->dd_reserved, ==, 0);
for (t = 0; t < DD_USED_NUM; t++)
ASSERT3U(dd->dd_phys->dd_used_breakdown[t], ==, 0);
VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
VERIFY(0 == zap_remove(mos,
dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
obj = dd->dd_object;
dsl_dir_close(dd, tag);
VERIFY(0 == dmu_object_free(mos, obj, tx));
}
int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj,
const struct lu_buf *buf, const char *name, int fl,
struct osd_thandle *oh)
{
dmu_buf_impl_t *db;
uchar_t *nv_value;
size_t size;
int nv_size;
int rc;
int too_big = 0;
LASSERT(obj->oo_sa_hdl);
if (obj->oo_sa_xattr == NULL) {
rc = __osd_xattr_cache(env, obj);
if (rc)
return rc;
}
LASSERT(obj->oo_sa_xattr);
/* Limited to 32k to keep nvpair memory allocations small */
if (buf->lb_len > DXATTR_MAX_ENTRY_SIZE) {
too_big = 1;
} else {
/* Prevent the DXATTR SA from consuming the entire SA
* region */
rc = -nvlist_size(obj->oo_sa_xattr, &size, NV_ENCODE_XDR);
if (rc)
return rc;
if (size + buf->lb_len > DXATTR_MAX_SA_SIZE)
too_big = 1;
}
/* even in case of -EFBIG we must lookup xattr and check can we
* rewrite it then delete from SA */
rc = -nvlist_lookup_byte_array(obj->oo_sa_xattr, name, &nv_value,
&nv_size);
if (rc == 0) {
if (fl & LU_XATTR_CREATE) {
return -EEXIST;
} else if (too_big) {
rc = -nvlist_remove(obj->oo_sa_xattr, name,
DATA_TYPE_BYTE_ARRAY);
if (rc < 0)
return rc;
rc = __osd_sa_xattr_schedule_update(env, obj, oh);
return rc == 0 ? -EFBIG : rc;
}
} else if (rc == -ENOENT) {
if (fl & LU_XATTR_REPLACE)
return -ENODATA;
else if (too_big)
return -EFBIG;
} else {
return rc;
}
/* Ensure xattr doesn't exist in ZAP */
if (obj->oo_xattr != ZFS_NO_OBJECT) {
struct osd_device *osd = osd_obj2dev(obj);
uint64_t objid;
rc = -zap_lookup(osd->od_os, obj->oo_xattr,
name, 8, 1, &objid);
if (rc == 0) {
rc = -dmu_object_free(osd->od_os, objid, oh->ot_tx);
if (rc == 0)
zap_remove(osd->od_os, obj->oo_xattr,
name, oh->ot_tx);
}
}
rc = -nvlist_add_byte_array(obj->oo_sa_xattr, name,
(uchar_t *)buf->lb_buf, buf->lb_len);
if (rc)
return rc;
/* batch updates only for just created dnodes where we
* used to set number of EAs in a single transaction */
db = (dmu_buf_impl_t *)obj->oo_db;
if (DB_DNODE(db)->dn_allocated_txg == oh->ot_tx->tx_txg)
rc = __osd_sa_xattr_schedule_update(env, obj, oh);
else
rc = __osd_sa_xattr_update(env, obj, oh);
return rc;
}
void
zfs_sa_upgrade(sa_handle_t *hdl, dmu_tx_t *tx)
{
dmu_buf_t *db = sa_get_db(hdl);
znode_t *zp = sa_get_userdata(hdl);
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
sa_bulk_attr_t bulk[20];
int count = 0;
sa_bulk_attr_t sa_attrs[20] = { { 0 } };
zfs_acl_locator_cb_t locate = { 0 };
uint64_t uid, gid, mode, rdev, xattr, parent;
uint64_t crtime[2], mtime[2], ctime[2];
zfs_acl_phys_t znode_acl;
char scanstamp[AV_SCANSTAMP_SZ];
boolean_t drop_lock = B_FALSE;
/*
* No upgrade if ACL isn't cached
* since we won't know which locks are held
* and ready the ACL would require special "locked"
* interfaces that would be messy
*/
if (zp->z_acl_cached == NULL || vnode_islnk(ZTOV(zp)))
return;
/*
* If the z_lock is held and we aren't the owner
* the just return since we don't want to deadlock
* trying to update the status of z_is_sa. This
* file can then be upgraded at a later time.
*
* Otherwise, we know we are doing the
* sa_update() that caused us to enter this function.
*/
if (mutex_owner(&zp->z_lock) != curthread) {
if (mutex_tryenter(&zp->z_lock) == 0)
return;
else
drop_lock = B_TRUE;
}
/* First do a bulk query of the attributes that aren't cached */
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL, &xattr, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
&znode_acl, 88);
if (sa_bulk_lookup_locked(hdl, bulk, count) != 0)
goto done;
/*
* While the order here doesn't matter its best to try and organize
* it is such a way to pick up an already existing layout number
*/
count = 0;
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
&zp->z_size, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GEN(zfsvfs),
NULL, &zp->z_gen, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PARENT(zfsvfs),
NULL, &parent, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
&zp->z_pflags, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_ATIME(zfsvfs), NULL,
zp->z_atime, 16);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MTIME(zfsvfs), NULL,
&mtime, 16);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CTIME(zfsvfs), NULL,
&ctime, 16);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
&crtime, 16);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_LINKS(zfsvfs), NULL,
&zp->z_links, 8);
if (vnode_isblk(zp->z_vnode) || vnode_islnk(zp->z_vnode))
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
&rdev, 8);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
&zp->z_acl_cached->z_acl_count, 8);
if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
zfs_acl_xform(zp, zp->z_acl_cached, CRED());
locate.cb_aclp = zp->z_acl_cached;
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
zfs_acl_data_locator, &locate, zp->z_acl_cached->z_acl_bytes);
if (xattr)
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_XATTR(zfsvfs),
NULL, &xattr, 8);
/* if scanstamp then add scanstamp */
if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
scanstamp, AV_SCANSTAMP_SZ);
SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SCANSTAMP(zfsvfs),
NULL, scanstamp, AV_SCANSTAMP_SZ);
zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
}
VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
VERIFY(sa_replace_all_by_template_locked(hdl, sa_attrs,
count, tx) == 0);
if (znode_acl.z_acl_extern_obj)
VERIFY(0 == dmu_object_free(zfsvfs->z_os,
znode_acl.z_acl_extern_obj, tx));
zp->z_is_sa = B_TRUE;
done:
if (drop_lock)
mutex_exit(&zp->z_lock);
}
/* ARGSUSED */
static void
dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx)
{
static const char *old_names[] = {
"scrub_bookmark",
"scrub_ddt_bookmark",
"scrub_ddt_class_max",
"scrub_queue",
"scrub_min_txg",
"scrub_max_txg",
"scrub_func",
"scrub_errors",
NULL
};
dsl_pool_t *dp = scn->scn_dp;
spa_t *spa = dp->dp_spa;
int i;
/* Remove any remnants of an old-style scrub. */
for (i = 0; old_names[i]; i++) {
(void) zap_remove(dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx);
}
if (scn->scn_phys.scn_queue_obj != 0) {
VERIFY(0 == dmu_object_free(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, tx));
scn->scn_phys.scn_queue_obj = 0;
}
/*
* If we were "restarted" from a stopped state, don't bother
* with anything else.
*/
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
if (complete)
scn->scn_phys.scn_state = DSS_FINISHED;
else
scn->scn_phys.scn_state = DSS_CANCELED;
spa_history_log_internal(spa, "scan done", tx,
"complete=%u", complete);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
mutex_enter(&spa->spa_scrub_lock);
while (spa->spa_scrub_inflight > 0) {
cv_wait(&spa->spa_scrub_io_cv,
&spa->spa_scrub_lock);
}
mutex_exit(&spa->spa_scrub_lock);
spa->spa_scrub_started = B_FALSE;
spa->spa_scrub_active = B_FALSE;
/*
* If the scrub/resilver completed, update all DTLs to
* reflect this. Whether it succeeded or not, vacate
* all temporary scrub DTLs.
*/
vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg,
complete ? scn->scn_phys.scn_max_txg : 0, B_TRUE);
if (complete) {
spa_event_notify(spa, NULL, scn->scn_phys.scn_min_txg ?
ESC_ZFS_RESILVER_FINISH : ESC_ZFS_SCRUB_FINISH);
}
spa_errlog_rotate(spa);
/*
* We may have finished replacing a device.
* Let the async thread assess this and handle the detach.
*/
spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
}
scn->scn_phys.scn_end_time = gethrestime_sec();
}
/* ARGSUSED */
static void
dsl_pool_scrub_cancel_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
{
dsl_pool_t *dp = arg1;
boolean_t *completep = arg2;
if (dp->dp_scrub_func == SCRUB_FUNC_NONE)
return;
mutex_enter(&dp->dp_scrub_cancel_lock);
if (dp->dp_scrub_restart) {
dp->dp_scrub_restart = B_FALSE;
*completep = B_FALSE;
}
/* XXX this is scrub-clean specific */
mutex_enter(&dp->dp_spa->spa_scrub_lock);
while (dp->dp_spa->spa_scrub_inflight > 0) {
cv_wait(&dp->dp_spa->spa_scrub_io_cv,
&dp->dp_spa->spa_scrub_lock);
}
mutex_exit(&dp->dp_spa->spa_scrub_lock);
dp->dp_spa->spa_scrub_active = B_FALSE;
dp->dp_scrub_func = SCRUB_FUNC_NONE;
VERIFY(0 == dmu_object_free(dp->dp_meta_objset,
dp->dp_scrub_queue_obj, tx));
dp->dp_scrub_queue_obj = 0;
bzero(&dp->dp_scrub_bookmark, sizeof (zbookmark_t));
bzero(&dp->dp_scrub_ddt_bookmark, sizeof (ddt_bookmark_t));
VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCRUB_QUEUE, tx));
VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCRUB_MIN_TXG, tx));
VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCRUB_MAX_TXG, tx));
VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCRUB_BOOKMARK, tx));
VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCRUB_FUNC, tx));
VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCRUB_ERRORS, tx));
(void) zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCRUB_DDT_BOOKMARK, tx);
(void) zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCRUB_DDT_CLASS_MAX, tx);
spa_history_internal_log(LOG_POOL_SCRUB_DONE, dp->dp_spa, tx, cr,
"complete=%u", *completep);
/* below is scrub-clean specific */
vdev_scrub_stat_update(dp->dp_spa->spa_root_vdev, POOL_SCRUB_NONE,
*completep);
/*
* If the scrub/resilver completed, update all DTLs to reflect this.
* Whether it succeeded or not, vacate all temporary scrub DTLs.
*/
vdev_dtl_reassess(dp->dp_spa->spa_root_vdev, tx->tx_txg,
*completep ? dp->dp_scrub_max_txg : 0, B_TRUE);
dp->dp_spa->spa_scrub_started = B_FALSE;
if (*completep)
spa_event_notify(dp->dp_spa, NULL, dp->dp_scrub_min_txg ?
ESC_ZFS_RESILVER_FINISH : ESC_ZFS_SCRUB_FINISH);
spa_errlog_rotate(dp->dp_spa);
/*
* We may have finished replacing a device.
* Let the async thread assess this and handle the detach.
*/
spa_async_request(dp->dp_spa, SPA_ASYNC_RESILVER_DONE);
dp->dp_scrub_min_txg = dp->dp_scrub_max_txg = 0;
mutex_exit(&dp->dp_scrub_cancel_lock);
}
void
vdev_indirect_births_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
{
VERIFY0(dmu_object_free(os, object, tx));
}
static int osd_object_destroy(const struct lu_env *env,
struct dt_object *dt, struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
char *buf = info->oti_str;
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
struct osd_thandle *oh;
int rc;
uint64_t oid, zapid;
ENTRY;
down_write(&obj->oo_guard);
if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
GOTO(out, rc = -ENOENT);
LASSERT(obj->oo_db != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
LASSERT(oh != NULL);
LASSERT(oh->ot_tx != NULL);
/* remove obj ref from index dir (it depends) */
zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
if (rc) {
CERROR("%s: zap_remove(%s) failed: rc = %d\n",
osd->od_svname, buf, rc);
GOTO(out, rc);
}
rc = osd_xattrs_destroy(env, obj, oh);
if (rc) {
CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
osd->od_svname, buf, rc);
GOTO(out, rc);
}
/* Remove object from inode accounting. It is not fatal for the destroy
* operation if something goes wrong while updating accounting, but we
* still log an error message to notify the administrator */
rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
obj->oo_attr.la_uid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
" %d: rc = %d\n", osd->od_svname, PFID(fid),
obj->oo_attr.la_uid, rc);
rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
obj->oo_attr.la_gid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
" %d: rc = %d\n", osd->od_svname, PFID(fid),
obj->oo_attr.la_gid, rc);
oid = obj->oo_db->db_object;
if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
/* this may happen if the destroy wasn't declared
* e.g. when the object is created and then destroyed
* in the same transaction - we don't need additional
* space for destroy specifically */
LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
if (rc)
CERROR("%s: failed to free %s %llu: rc = %d\n",
osd->od_svname, buf, oid, rc);
} else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
if (rc)
CERROR("%s: failed to free %s %llu: rc = %d\n",
osd->od_svname, buf, oid, rc);
} else { /* asynchronous destroy */
rc = osd_object_unlinked_add(obj, oh);
if (rc)
GOTO(out, rc);
rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
oid, oh->ot_tx);
if (rc)
CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
osd->od_svname, buf, oid, rc);
}
out:
/* not needed in the cache anymore */
set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
if (rc == 0)
obj->oo_destroyed = 1;
up_write(&obj->oo_guard);
RETURN (0);
}
/*
* 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);
}
void
bplist_destroy(objset_t *mos, uint64_t object, dmu_tx_t *tx)
{
VERIFY(dmu_object_free(mos, object, tx) == 0);
}
static int osd_object_destroy(const struct lu_env *env,
struct dt_object *dt, struct thandle *th)
{
char *buf = osd_oti_get(env)->oti_str;
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
struct osd_thandle *oh;
int rc;
uint64_t oid, zapid;
ENTRY;
LASSERT(obj->oo_db != NULL);
LASSERT(dt_object_exists(dt));
LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
oh = container_of0(th, struct osd_thandle, ot_super);
LASSERT(oh != NULL);
LASSERT(oh->ot_tx != NULL);
/* remove obj ref from index dir (it depends) */
zapid = osd_get_name_n_idx(env, osd, fid, buf);
rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
if (rc) {
CERROR("%s: zap_remove(%s) failed: rc = %d\n",
osd->od_svname, buf, rc);
GOTO(out, rc);
}
rc = osd_xattrs_destroy(env, obj, oh);
if (rc) {
CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
osd->od_svname, buf, rc);
GOTO(out, rc);
}
/* Remove object from inode accounting. It is not fatal for the destroy
* operation if something goes wrong while updating accounting, but we
* still log an error message to notify the administrator */
rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
obj->oo_attr.la_uid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
" %d: rc = %d\n", osd->od_svname, PFID(fid),
obj->oo_attr.la_uid, rc);
rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
obj->oo_attr.la_gid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
" %d: rc = %d\n", osd->od_svname, PFID(fid),
obj->oo_attr.la_gid, rc);
oid = obj->oo_db->db_object;
if (obj->oo_destroy == OSD_DESTROY_SYNC) {
rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
if (rc)
CERROR("%s: failed to free %s "LPU64": rc = %d\n",
osd->od_svname, buf, oid, rc);
} else { /* asynchronous destroy */
rc = osd_object_unlinked_add(obj, oh);
if (rc)
GOTO(out, rc);
rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
oid, oh->ot_tx);
if (rc)
CERROR("%s: zap_add_int() failed %s "LPU64": rc = %d\n",
osd->od_svname, buf, oid, rc);
}
out:
/* not needed in the cache anymore */
set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
if (rc == 0)
obj->oo_destroyed = 1;
RETURN (0);
}
