static void
dsl_deleg_unset_sync(void *arg1, void *arg2, dmu_tx_t *tx)
{
dsl_dir_t *dd = arg1;
nvlist_t *nvp = arg2;
objset_t *mos = dd->dd_pool->dp_meta_objset;
nvpair_t *whopair = NULL;
uint64_t zapobj = dd->dd_phys->dd_deleg_zapobj;
if (zapobj == 0)
return;
while (whopair = nvlist_next_nvpair(nvp, whopair)) {
const char *whokey = nvpair_name(whopair);
nvlist_t *perms;
nvpair_t *permpair = NULL;
uint64_t jumpobj;
if (nvpair_value_nvlist(whopair, &perms) != 0) {
if (zap_lookup(mos, zapobj, whokey, 8,
1, &jumpobj) == 0) {
(void) zap_remove(mos, zapobj, whokey, tx);
VERIFY(0 == zap_destroy(mos, jumpobj, tx));
}
spa_history_log_internal(LOG_DS_PERM_WHO_REMOVE,
dd->dd_pool->dp_spa, tx,
"%s dataset = %llu", whokey,
dd->dd_phys->dd_head_dataset_obj);
continue;
}
if (zap_lookup(mos, zapobj, whokey, 8, 1, &jumpobj) != 0)
continue;
while (permpair = nvlist_next_nvpair(perms, permpair)) {
const char *perm = nvpair_name(permpair);
uint64_t n = 0;
(void) zap_remove(mos, jumpobj, perm, tx);
if (zap_count(mos, jumpobj, &n) == 0 && n == 0) {
(void) zap_remove(mos, zapobj,
whokey, tx);
VERIFY(0 == zap_destroy(mos,
jumpobj, tx));
}
spa_history_log_internal(LOG_DS_PERM_REMOVE,
dd->dd_pool->dp_spa, tx,
"%s %s dataset = %llu", whokey, perm,
dd->dd_phys->dd_head_dataset_obj);
}
}
}
static void
vdev_initialize_change_state(vdev_t *vd, vdev_initializing_state_t new_state)
{
ASSERT(MUTEX_HELD(&vd->vdev_initialize_lock));
spa_t *spa = vd->vdev_spa;
if (new_state == vd->vdev_initialize_state)
return;
/*
* Copy the vd's guid, this will be freed by the sync task.
*/
uint64_t *guid = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
*guid = vd->vdev_guid;
/*
* If we're suspending, then preserving the original start time.
*/
if (vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED) {
vd->vdev_initialize_action_time = gethrestime_sec();
}
vd->vdev_initialize_state = new_state;
dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
dsl_sync_task_nowait(spa_get_dsl(spa), vdev_initialize_zap_update_sync,
guid, 2, ZFS_SPACE_CHECK_RESERVED, tx);
switch (new_state) {
case VDEV_INITIALIZE_ACTIVE:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s activated", vd->vdev_path);
break;
case VDEV_INITIALIZE_SUSPENDED:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s suspended", vd->vdev_path);
break;
case VDEV_INITIALIZE_CANCELED:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s canceled", vd->vdev_path);
break;
case VDEV_INITIALIZE_COMPLETE:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s complete", vd->vdev_path);
break;
default:
panic("invalid state %llu", (unsigned long long)new_state);
}
dmu_tx_commit(tx);
}
void
spa_history_log_version(spa_t *spa, const char *operation)
{
spa_history_log_internal(spa, operation, NULL,
"pool version %llu; software version %llu/%d; uts %s %s %s %s",
(u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
utsname.nodename, utsname.release, utsname.version,
utsname.machine);
}
static void
feature_decr_sync(void *arg, dmu_tx_t *tx)
{
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
zfeature_info_t *feature = arg;
spa_feature_decr(spa, feature, tx);
spa_history_log_internal(spa, "zhack feature decr", tx,
"name=%s", feature->fi_guid);
}
void
spa_history_log_version(spa_t *spa, const char *operation, dmu_tx_t *tx)
{
utsname_t *u = utsname();
spa_history_log_internal(spa, operation, tx,
"pool version %llu; software version %llu/%llu; uts %s %s %s %s",
(u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
u->nodename, u->release, u->version, u->machine);
}
static void
feature_enable_sync(void *arg, dmu_tx_t *tx)
{
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
zfeature_info_t *feature = arg;
spa_feature_enable(spa, feature, tx);
spa_history_log_internal(spa, "zhack enable feature", tx,
"name=%s can_readonly=%u",
feature->fi_guid, feature->fi_can_readonly);
}
static void
zhack_feature_enable_sync(void *arg, dmu_tx_t *tx)
{
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
zfeature_info_t *feature = arg;
feature_enable_sync(spa, feature, tx);
spa_history_log_internal(spa, "zhack enable feature", tx,
"name=%s flags=%u",
feature->fi_guid, feature->fi_flags);
}
static void
feature_decr_sync(void *arg, dmu_tx_t *tx)
{
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
zfeature_info_t *feature = arg;
uint64_t refcount;
VERIFY0(feature_get_refcount_from_disk(spa, feature, &refcount));
feature_sync(spa, feature, refcount - 1, tx);
spa_history_log_internal(spa, "zhack feature decr", tx,
"name=%s", feature->fi_guid);
}
void
spa_history_log_version(spa_t *spa, const char *operation)
{
#ifdef _KERNEL
uint64_t current_vers = spa_version(spa);
spa_history_log_internal(spa, operation, NULL,
"pool version %llu; software version %llu/%d; uts %s %s %s %s",
(u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
utsname.nodename, utsname.release, utsname.version,
utsname.machine);
cmn_err(CE_CONT, "!%s version %llu pool %s using %llu", operation,
(u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
#endif
}
static void
dsl_crypto_key_new_sync(void *arg1, dmu_tx_t *tx)
{
struct knarg *kn = arg1;
dsl_dataset_t *ds = kn->kn_ds;
/*
* Generate a new key and add it to the keychain to be valid from
* this txg onwards.
*/
dsl_keychain_set_key(ds->ds_dir, tx, kn->kn_wkeybuf, kn->kn_wkeylen,
gethrestime_sec());
zcrypt_keychain_insert(&kn->kn_skn->skn_keychain,
tx->tx_txg, kn->kn_txgkey);
spa_history_log_internal(
dsl_dataset_get_spa(ds), "key create", tx,
"rekey succeeded dataset = %llu", ds->ds_object);
}
void
spa_history_log_version(spa_t *spa, history_internal_events_t event)
{
#ifdef _KERNEL
uint64_t current_vers = spa_version(spa);
if (current_vers >= SPA_VERSION_ZPOOL_HISTORY) {
spa_history_log_internal(event, spa, NULL,
"pool spa %llu; zfs spa %llu; zpl %d; uts %s %s %s %s",
(u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
utsname.nodename, utsname.release, utsname.version,
utsname.machine);
}
cmn_err(CE_CONT, "!%s version %llu pool %s using %llu",
event == LOG_POOL_IMPORT ? "imported" :
event == LOG_POOL_CREATE ? "created" : "accessed",
(u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
#endif
}
static void
dsl_deleg_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
{
dsl_dir_t *dd = arg1;
nvlist_t *nvp = arg2;
objset_t *mos = dd->dd_pool->dp_meta_objset;
nvpair_t *whopair = NULL;
uint64_t zapobj = dd->dd_phys->dd_deleg_zapobj;
if (zapobj == 0) {
dmu_buf_will_dirty(dd->dd_dbuf, tx);
zapobj = dd->dd_phys->dd_deleg_zapobj = zap_create(mos,
DMU_OT_DSL_PERMS, DMU_OT_NONE, 0, tx);
}
while (whopair = nvlist_next_nvpair(nvp, whopair)) {
const char *whokey = nvpair_name(whopair);
nvlist_t *perms;
nvpair_t *permpair = NULL;
uint64_t jumpobj;
VERIFY(nvpair_value_nvlist(whopair, &perms) == 0);
if (zap_lookup(mos, zapobj, whokey, 8, 1, &jumpobj) != 0) {
jumpobj = zap_create(mos, DMU_OT_DSL_PERMS,
DMU_OT_NONE, 0, tx);
VERIFY(zap_update(mos, zapobj,
whokey, 8, 1, &jumpobj, tx) == 0);
}
while (permpair = nvlist_next_nvpair(perms, permpair)) {
const char *perm = nvpair_name(permpair);
uint64_t n = 0;
VERIFY(zap_update(mos, jumpobj,
perm, 8, 1, &n, tx) == 0);
spa_history_log_internal(LOG_DS_PERM_UPDATE,
dd->dd_pool->dp_spa, tx,
"%s %s dataset = %llu", whokey, perm,
dd->dd_phys->dd_head_dataset_obj);
}
}
}
/*
* dsl_crypto_key_change
*
* The old key must already be present in memory since the user interface
* doesn't provide away to prompt or retrieve the old key.
*/
static void
dsl_crypto_key_change_sync(void *arg1, dmu_tx_t *tx)
{
struct wkey_change_arg *ca = arg1;
dsl_dataset_t *ds = ca->ca_ds;
size_t wkeylen;
char *wkeybuf = NULL;
zcrypt_key_t *txgkey;
zap_cursor_t zc;
zap_attribute_t za;
objset_t *mos;
uint64_t keychain_zapobj;
spa_t *spa;
zcrypt_keystore_node_t *zkn;
ASSERT(RRW_WRITE_HELD(&ds->ds_dir->dd_pool->dp_config_rwlock));
mos = ds->ds_dir->dd_pool->dp_meta_objset;
keychain_zapobj = dsl_dir_phys(ds->ds_dir)->dd_keychain_obj;
/*
* To allow for the case were the keychains of child datasets
* are not loaded (ie an explicit 'zfs key -u tank/fs/sub' had
* been done some time before doing 'zfs key -c tank/fs') we itterate
* over the zap objects on disk rather than copying from the
* in memory keystore node.
*/
for (zap_cursor_init(&zc, mos, keychain_zapobj);
zap_cursor_retrieve(&zc, &za) == 0;
zap_cursor_advance(&zc)) {
wkeylen = za.za_num_integers;
wkeybuf = kmem_alloc(wkeylen, KM_SLEEP);
VERIFY(zap_lookup_uint64(mos, keychain_zapobj,
(uint64_t *)za.za_name, 1, 1, wkeylen, wkeybuf) == 0);
VERIFY(zcrypt_unwrap_key(ca->ca_old_key,
ds->ds_objset->os_crypt, wkeybuf, wkeylen, &txgkey) == 0);
kmem_free(wkeybuf, wkeylen);
VERIFY(zcrypt_wrap_key(ca->ca_new_key, txgkey,
&wkeybuf, &wkeylen,
zio_crypt_select_wrap(ds->ds_objset->os_crypt)) == 0);
zcrypt_key_free(txgkey);
VERIFY(zap_update_uint64(mos, keychain_zapobj,
(uint64_t *)za.za_name, 1, 1, wkeylen, wkeybuf, tx) == 0);
kmem_free(wkeybuf, wkeylen);
}
zap_cursor_fini(&zc);
spa = dsl_dataset_get_spa(ds);
/*
* If the wrapping key is loaded switch the in memory copy now.
*/
zkn = zcrypt_keystore_find_node(spa, ds->ds_object, B_FALSE);
if (zkn != NULL) {
zcrypt_key_free(zkn->skn_wrapkey);
zkn->skn_wrapkey = zcrypt_key_copy(ca->ca_new_key);
}
spa_history_log_internal(spa, "key change", tx,
"succeeded dataset = %llu", ds->ds_object);
}
/*
* dsl_crypto_key_clone
*
* Our caller (dmu_objset_create_sync) must have a lock on the dataset.
*/
int
dsl_crypto_key_clone(dsl_dir_t *dd, dsl_dataset_phys_t *dsphys,
uint64_t dsobj, dsl_dataset_t *clone_origin,
dsl_crypto_ctx_t *ctx, dmu_tx_t *tx)
{
zcrypt_key_t *txgkey;
zcrypt_key_t *wrappingkey = NULL;
dsl_dataset_t *origin;
uint64_t crypt;
spa_t *spa = dd->dd_pool->dp_spa;
int error = 0;
ASSERT(ctx != NULL);
if (BP_IS_HOLE(&dsphys->ds_bp)) {
origin = ctx->dcc_origin_ds;
} else {
origin = clone_origin;
}
ASSERT(origin != NULL);
/*
* Need to look at the value of crypt on the origin snapshot
* since it is sticky for encryption.
*/
VERIFY(dsl_prop_get_ds(origin, zfs_prop_to_name(ZFS_PROP_ENCRYPTION),
8, 1, &crypt,/* DSL_PROP_GET_EFFECTIVE,*/ NULL) == 0);
if (crypt == ZIO_CRYPT_OFF) {
return (0);
}
wrappingkey = ctx->dcc_wrap_key;
dsl_keychain_create_obj(dd, tx);
dsl_keychain_clone_phys(origin, dd, tx, wrappingkey);
if (ctx->dcc_salt != 0) {
objset_t *mos = dd->dd_pool->dp_meta_objset;
uint64_t props_zapobj = dsl_dir_phys(dd)->dd_props_zapobj;
error = zap_update(mos, props_zapobj,
zfs_prop_to_name(ZFS_PROP_SALT), 8, 1,
(void *)&ctx->dcc_salt, tx);
}
if (ctx->dcc_clone_newkey) {
caddr_t wkeybuf;
size_t wkeylen;
/*
* Generate a new key and add it to the keychain
* to be valid from this txg onwards.
*/
zcrypt_key_hold(wrappingkey, FTAG);
txgkey = zcrypt_key_gen(crypt);
VERIFY(zcrypt_wrap_key(wrappingkey, txgkey,
&wkeybuf, &wkeylen, zio_crypt_select_wrap(crypt)) == 0);
zcrypt_key_release(wrappingkey, FTAG);
dsl_keychain_set_key(dd, tx, wkeybuf, wkeylen,
dsphys->ds_creation_time);
kmem_free(wkeybuf, wkeylen);
zcrypt_key_free(txgkey);
spa_history_log_internal(spa, "key create", tx,
"rekey succeeded dataset = %llu from dataset = %llu",
dsobj, clone_origin->ds_object);
} else {
spa_history_log_internal(spa, "key create", tx,
"succeeded dataset = %llu from dataset = %llu",
dsobj, clone_origin->ds_object);
}
if (wrappingkey != NULL) {
error = dsl_keychain_load_dd(dd, dsobj, crypt, wrappingkey);
}
return (error);
}
/*
* dsl_crypto_key_gen - Generate dataset key
*
* Generate a new key for this dataset based on its encryption property type.
* Store the key as a usable zcrypt_key_t in the in memory keystore and
* put the wrapped version of it in the on disk keychain.
*
* returns 0 on success
*/
int
dsl_crypto_key_create(dsl_dir_t *dd, dsl_dataset_phys_t *dsphys,
uint64_t dsobj, dsl_crypto_ctx_t *ctx, dmu_tx_t *tx)
{
int error = -1;
#ifdef DEBUG
zcrypt_key_t *debugkey;
#endif /* DEBUG */
zcrypt_key_t *wrappingkey, *dslkey;
caddr_t wkeybuf = NULL;
size_t wkeylen = 0;
uint64_t crypt;
spa_t *spa = dd->dd_pool->dp_spa;
zcrypt_keystore_node_t *skn;
if (ctx == NULL) {
return (0);
}
crypt = ctx->dcc_crypt;
if (crypt == ZIO_CRYPT_OFF || crypt == ZIO_CRYPT_INHERIT) {
return (0);
}
crypt = zio_crypt_select(crypt, ZIO_CRYPT_ON_VALUE);
wrappingkey = ctx->dcc_wrap_key;
skn = zcrypt_keystore_insert(spa, dsobj, wrappingkey);
if (skn == NULL) {
error = ENOKEY;
goto out;
}
dslkey = zcrypt_key_gen(crypt);
zcrypt_key_hold(wrappingkey, FTAG);
error = zcrypt_wrap_key(wrappingkey, dslkey,
&wkeybuf, &wkeylen, zio_crypt_select_wrap(crypt));
if (error != 0) {
zcrypt_key_free(dslkey);
spa_history_log_internal(spa, "key create", tx,
"failed dataset = %llu unable to wrap key", dsobj, error);
goto out;
}
dsl_keychain_create_obj(dd, tx);
dsl_keychain_set_key(dd, tx, wkeybuf, wkeylen,
dsphys->ds_creation_time);
zcrypt_keychain_insert(&skn->skn_keychain,
dsphys->ds_creation_txg, dslkey);
if (ctx->dcc_salt != 0) {
objset_t *mos = dd->dd_pool->dp_meta_objset;
uint64_t props_zapobj = dsl_dir_phys(dd)->dd_props_zapobj;
error = zap_update(mos, props_zapobj,
zfs_prop_to_name(ZFS_PROP_SALT), 8, 1,
(void *)&ctx->dcc_salt, tx);
}
if (error == 0)
spa_history_log_internal(spa, "key create", tx,
"succeeded dataset = %llu", dsobj);
#ifdef DEBUG
ASSERT3U(zcrypt_unwrap_key(wrappingkey, crypt,
wkeybuf, wkeylen, &debugkey), ==, 0);
zcrypt_key_compare(dslkey, debugkey);
zcrypt_key_free(debugkey);
#endif /* DEBUG */
out:
zcrypt_key_release(wrappingkey, FTAG);
if (wkeylen > 0) {
kmem_free(wkeybuf, wkeylen);
}
return (error);
}
/* 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();
}
static void
dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
pool_scan_func_t *funcp = arg;
dmu_object_type_t ot = 0;
dsl_pool_t *dp = scn->scn_dp;
spa_t *spa = dp->dp_spa;
ASSERT(scn->scn_phys.scn_state != DSS_SCANNING);
ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS);
bzero(&scn->scn_phys, sizeof (scn->scn_phys));
scn->scn_phys.scn_func = *funcp;
scn->scn_phys.scn_state = DSS_SCANNING;
scn->scn_phys.scn_min_txg = 0;
scn->scn_phys.scn_max_txg = tx->tx_txg;
scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */
scn->scn_phys.scn_start_time = gethrestime_sec();
scn->scn_phys.scn_errors = 0;
scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc;
scn->scn_restart_txg = 0;
spa_scan_stat_init(spa);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max;
/* rewrite all disk labels */
vdev_config_dirty(spa->spa_root_vdev);
if (vdev_resilver_needed(spa->spa_root_vdev,
&scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) {
spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START);
} else {
spa_event_notify(spa, NULL, ESC_ZFS_SCRUB_START);
}
spa->spa_scrub_started = B_TRUE;
/*
* If this is an incremental scrub, limit the DDT scrub phase
* to just the auto-ditto class (for correctness); the rest
* of the scrub should go faster using top-down pruning.
*/
if (scn->scn_phys.scn_min_txg > TXG_INITIAL)
scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO;
}
/* back to the generic stuff */
if (dp->dp_blkstats == NULL) {
dp->dp_blkstats =
kmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP);
}
bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
if (spa_version(spa) < SPA_VERSION_DSL_SCRUB)
ot = DMU_OT_ZAP_OTHER;
scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset,
ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx);
dsl_scan_sync_state(scn, tx);
spa_history_log_internal(spa, "scan setup", tx,
"func=%u mintxg=%llu maxtxg=%llu",
*funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg);
}
