int
spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
{
int err = 0;
dmu_tx_t *tx;
nvlist_t *nvarg;
if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa))
return (SET_ERROR(EINVAL));
tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
err = dmu_tx_assign(tx, TXG_WAIT);
if (err) {
dmu_tx_abort(tx);
return (err);
}
VERIFY0(nvlist_dup(nvl, &nvarg, KM_SLEEP));
if (spa_history_zone() != NULL) {
fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
spa_history_zone());
}
fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
/* Kick this off asynchronously; errors are ignored. */
dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
nvarg, 0, ZFS_SPACE_CHECK_NONE, tx);
dmu_tx_commit(tx);
/* spa_history_log_sync will free nvl */
return (err);
}
/*
* Write out a history event.
*/
int
spa_history_log(spa_t *spa, const char *history_str, history_log_type_t what)
{
history_arg_t *ha;
int err = 0;
dmu_tx_t *tx;
ASSERT(what != LOG_INTERNAL);
tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
err = dmu_tx_assign(tx, TXG_WAIT);
if (err) {
dmu_tx_abort(tx);
return (err);
}
ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
ha->ha_history_str = strdup(history_str);
ha->ha_zone = strdup(spa_history_zone());
ha->ha_log_type = what;
ha->ha_uid = crgetuid(CRED());
/* Kick this off asynchronously; errors are ignored. */
dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
spa_history_log_sync, spa, ha, 0, tx);
dmu_tx_commit(tx);
/* spa_history_log_sync will free ha and strings */
return (err);
}
void
spa_history_log_internal(spa_t *spa, const char *operation,
dmu_tx_t *tx, const char *fmt, ...)
{
dmu_tx_t *htx = tx;
va_list adx;
nvlist_t *nvl;
/* create a tx if we didn't get one */
if (tx == NULL) {
if (!spa_get_dsl(spa)) return;
htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
dmu_tx_abort(htx);
return;
}
}
va_start(adx, fmt);
VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));
//log_internal(nvl, operation, spa, htx, fmt, adx);
va_end(adx);
/* if we didn't get a tx from the caller, commit the one we made */
if (tx == NULL)
dmu_tx_commit(htx);
}
/* Takes care of physical writing and limiting # of concurrent ZIOs. */
static int
vdev_initialize_write(vdev_t *vd, uint64_t start, uint64_t size, abd_t *data)
{
spa_t *spa = vd->vdev_spa;
/* Limit inflight initializing I/Os */
mutex_enter(&vd->vdev_initialize_io_lock);
while (vd->vdev_initialize_inflight >= zfs_initialize_limit) {
cv_wait(&vd->vdev_initialize_io_cv,
&vd->vdev_initialize_io_lock);
}
vd->vdev_initialize_inflight++;
mutex_exit(&vd->vdev_initialize_io_lock);
dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
uint64_t txg = dmu_tx_get_txg(tx);
spa_config_enter(spa, SCL_STATE_ALL, vd, RW_READER);
mutex_enter(&vd->vdev_initialize_lock);
if (vd->vdev_initialize_offset[txg & TXG_MASK] == 0) {
uint64_t *guid = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
*guid = vd->vdev_guid;
/* This is the first write of this txg. */
dsl_sync_task_nowait(spa_get_dsl(spa),
vdev_initialize_zap_update_sync, guid, 2,
ZFS_SPACE_CHECK_RESERVED, tx);
}
/*
* We know the vdev struct will still be around since all
* consumers of vdev_free must stop the initialization first.
*/
if (vdev_initialize_should_stop(vd)) {
mutex_enter(&vd->vdev_initialize_io_lock);
ASSERT3U(vd->vdev_initialize_inflight, >, 0);
vd->vdev_initialize_inflight--;
mutex_exit(&vd->vdev_initialize_io_lock);
spa_config_exit(vd->vdev_spa, SCL_STATE_ALL, vd);
mutex_exit(&vd->vdev_initialize_lock);
dmu_tx_commit(tx);
return (SET_ERROR(EINTR));
}
mutex_exit(&vd->vdev_initialize_lock);
vd->vdev_initialize_offset[txg & TXG_MASK] = start + size;
zio_nowait(zio_write_phys(spa->spa_txg_zio[txg & TXG_MASK], vd, start,
size, data, ZIO_CHECKSUM_OFF, vdev_initialize_cb, NULL,
ZIO_PRIORITY_INITIALIZING, ZIO_FLAG_CANFAIL, B_FALSE));
/* vdev_initialize_cb releases SCL_STATE_ALL */
dmu_tx_commit(tx);
return (0);
}
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_internal_log(history_internal_events_t event, spa_t *spa,
dmu_tx_t *tx, cred_t *cr, const char *fmt, ...)
{
history_arg_t *hap;
char *str;
va_list adx;
hap = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
str = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP);
va_start(adx, fmt);
(void) vsnprintf(str, HIS_MAX_RECORD_LEN, fmt, adx);
va_end(adx);
hap->ha_log_type = LOG_INTERNAL;
hap->ha_history_str = str;
hap->ha_event = event;
hap->ha_zone[0] = '\0';
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(spa, hap, cr, tx);
} else {
dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
spa_history_log_sync, spa, hap, 0, tx);
}
/* spa_history_log_sync() will free hap and str */
}
/*
* Open-context function to add one entry to the new mapping. The new
* entry will be remembered and written from syncing context.
*/
static void
spa_condense_indirect_commit_entry(spa_t *spa,
vdev_indirect_mapping_entry_phys_t *vimep, uint32_t count)
{
spa_condensing_indirect_t *sci = spa->spa_condensing_indirect;
ASSERT3U(count, <, DVA_GET_ASIZE(&vimep->vimep_dst));
dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
dmu_tx_hold_space(tx, sizeof (*vimep) + sizeof (count));
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
int txgoff = dmu_tx_get_txg(tx) & TXG_MASK;
/*
* If we are the first entry committed this txg, kick off the sync
* task to write to the MOS on our behalf.
*/
if (list_is_empty(&sci->sci_new_mapping_entries[txgoff])) {
dsl_sync_task_nowait(dmu_tx_pool(tx),
spa_condense_indirect_commit_sync, sci,
0, ZFS_SPACE_CHECK_NONE, tx);
}
vdev_indirect_mapping_entry_t *vime =
kmem_alloc(sizeof (*vime), KM_SLEEP);
vime->vime_mapping = *vimep;
vime->vime_obsolete_count = count;
list_insert_tail(&sci->sci_new_mapping_entries[txgoff], vime);
dmu_tx_commit(tx);
}
static void
dsl_dataset_user_release_onexit(void *arg)
{
zfs_hold_cleanup_arg_t *ca = arg;
spa_t *spa;
int error;
error = spa_open(ca->zhca_spaname, &spa, FTAG);
if (error != 0) {
zfs_dbgmsg("couldn't release holds on pool=%s "
"because pool is no longer loaded",
ca->zhca_spaname);
return;
}
if (spa_load_guid(spa) != ca->zhca_spa_load_guid) {
zfs_dbgmsg("couldn't release holds on pool=%s "
"because pool is no longer loaded (guid doesn't match)",
ca->zhca_spaname);
spa_close(spa, FTAG);
return;
}
(void) dsl_dataset_user_release_tmp(spa_get_dsl(spa), ca->zhca_holds);
fnvlist_free(ca->zhca_holds);
kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
spa_close(spa, FTAG);
}
static void
log_internal(history_internal_events_t event, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
history_arg_t *ha;
va_list adx_copy;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
*/
if (tx->tx_txg == TXG_INITIAL)
return;
ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
va_copy(adx_copy, adx);
ha->ha_history_str = kmem_vasprintf(fmt, adx_copy);
va_end(adx_copy);
ha->ha_log_type = LOG_INTERNAL;
ha->ha_event = event;
ha->ha_zone = NULL;
ha->ha_uid = 0;
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(spa, ha, tx);
} else {
dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
spa_history_log_sync, spa, ha, 0, tx);
}
/* spa_history_log_sync() will free ha and strings */
}
/*
* The nvlist will be consumed by this call.
*/
static void
log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
char *msg;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
* Likewise if the pool is not writeable.
*/
if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
fnvlist_free(nvl);
return;
}
msg = kmem_vasprintf(fmt, adx);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
strfree(msg);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(nvl, tx);
} else {
dsl_sync_task_nowait(spa_get_dsl(spa),
spa_history_log_sync, nvl, 0, ZFS_SPACE_CHECK_NONE, tx);
}
/* spa_history_log_sync() will free nvl */
}
int
dmu_objset_snapshot(char *fsname, char *snapname, boolean_t recursive)
{
dsl_sync_task_t *dst;
struct osnode *osn;
struct snaparg sn = { 0 };
spa_t *spa;
int err;
(void) strcpy(sn.failed, fsname);
err = spa_open(fsname, &spa, FTAG);
if (err)
return (err);
sn.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
sn.snapname = snapname;
list_create(&sn.objsets, sizeof (struct osnode),
offsetof(struct osnode, node));
if (recursive) {
sn.checkperms = B_TRUE;
err = dmu_objset_find(fsname,
dmu_objset_snapshot_one, &sn, DS_FIND_CHILDREN);
} else {
sn.checkperms = B_FALSE;
err = dmu_objset_snapshot_one(fsname, &sn);
}
if (err)
goto out;
err = dsl_sync_task_group_wait(sn.dstg);
for (dst = list_head(&sn.dstg->dstg_tasks); dst;
dst = list_next(&sn.dstg->dstg_tasks, dst)) {
dsl_dataset_t *ds = dst->dst_arg1;
if (dst->dst_err)
dsl_dataset_name(ds, sn.failed);
}
out:
while (osn = list_head(&sn.objsets)) {
list_remove(&sn.objsets, osn);
zil_resume(dmu_objset_zil(osn->os));
dmu_objset_close(osn->os);
kmem_free(osn, sizeof (struct osnode));
}
list_destroy(&sn.objsets);
if (err)
(void) strcpy(fsname, sn.failed);
dsl_sync_task_group_destroy(sn.dstg);
spa_close(spa, FTAG);
return (err);
}
dsl_pool_t *
dmu_objset_pool(objset_t *os)
{
dsl_dataset_t *ds;
if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
return (ds->ds_dir->dd_pool);
else
return (spa_get_dsl(os->os_spa));
}
/*
* Called from open context to perform a callback in syncing context. Waits
* for the operation to complete.
*
* The checkfunc will be called from open context as a preliminary check
* which can quickly fail. If it succeeds, it will be called again from
* syncing context. The checkfunc should generally be designed to work
* properly in either context, but if necessary it can check
* dmu_tx_is_syncing(tx).
*
* The synctask infrastructure enforces proper locking strategy with respect
* to the dp_config_rwlock -- the lock will always be held when the callbacks
* are called. It will be held for read during the open-context (preliminary)
* call to the checkfunc, and then held for write from syncing context during
* the calls to the check and sync funcs.
*
* A dataset or pool name can be passed as the first argument. Typically,
* the check func will hold, check the return value of the hold, and then
* release the dataset. The sync func will VERIFYO(hold()) the dataset.
* This is safe because no changes can be made between the check and sync funcs,
* and the sync func will only be called if the check func successfully opened
* the dataset.
*/
int
dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
dsl_syncfunc_t *syncfunc, void *arg,
int blocks_modified, zfs_space_check_t space_check)
{
spa_t *spa;
dmu_tx_t *tx;
int err;
dsl_sync_task_t dst = { { { NULL } } };
dsl_pool_t *dp;
err = spa_open(pool, &spa, FTAG);
if (err != 0)
return (err);
dp = spa_get_dsl(spa);
top:
tx = dmu_tx_create_dd(dp->dp_mos_dir);
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
dst.dst_pool = dp;
dst.dst_txg = dmu_tx_get_txg(tx);
dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT;
dst.dst_space_check = space_check;
dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc;
dst.dst_syncfunc = syncfunc;
dst.dst_arg = arg;
dst.dst_error = 0;
dst.dst_nowaiter = B_FALSE;
dsl_pool_config_enter(dp, FTAG);
err = dst.dst_checkfunc(arg, tx);
dsl_pool_config_exit(dp, FTAG);
if (err != 0) {
dmu_tx_commit(tx);
spa_close(spa, FTAG);
return (err);
}
VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, &dst, dst.dst_txg));
dmu_tx_commit(tx);
txg_wait_synced(dp, dst.dst_txg);
if (dst.dst_error == EAGAIN) {
txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE);
goto top;
}
spa_close(spa, FTAG);
return (dst.dst_error);
}
/*
* Write out a history event.
*/
int
spa_history_log(spa_t *spa, const char *history_str, history_log_type_t what)
{
history_arg_t ha;
ASSERT(what != LOG_INTERNAL);
ha.ha_history_str = history_str;
ha.ha_log_type = what;
(void) strlcpy(ha.ha_zone, spa_history_zone(), sizeof (ha.ha_zone));
return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_history_log_sync,
spa, &ha, 0));
}
int
dmu_objset_snapshot(char *fsname, char *snapname,
nvlist_t *props, boolean_t recursive)
{
dsl_sync_task_t *dst;
struct snaparg *sn;
spa_t *spa;
int err;
sn = kmem_alloc(sizeof (struct snaparg), KM_SLEEP);
(void) strcpy(sn->failed, fsname);
err = spa_open(fsname, &spa, FTAG);
if (err) {
kmem_free(sn, sizeof (struct snaparg));
return (err);
}
sn->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
sn->snapname = snapname;
sn->props = props;
if (recursive) {
sn->checkperms = B_TRUE;
err = dmu_objset_find(fsname,
dmu_objset_snapshot_one, sn, DS_FIND_CHILDREN);
} else {
sn->checkperms = B_FALSE;
err = dmu_objset_snapshot_one(fsname, sn);
}
if (err == 0)
err = dsl_sync_task_group_wait(sn->dstg);
for (dst = list_head(&sn->dstg->dstg_tasks); dst;
dst = list_next(&sn->dstg->dstg_tasks, dst)) {
objset_t *os = dst->dst_arg1;
dsl_dataset_t *ds = os->os->os_dsl_dataset;
if (dst->dst_err)
dsl_dataset_name(ds, sn->failed);
zil_resume(dmu_objset_zil(os));
dmu_objset_close(os);
}
if (err)
(void) strcpy(fsname, sn->failed);
dsl_sync_task_group_destroy(sn->dstg);
spa_close(spa, FTAG);
kmem_free(sn, sizeof (struct snaparg));
return (err);
}
static int
vdev_queue_max_async_writes(spa_t *spa)
{
int writes;
uint64_t dirty = 0;
dsl_pool_t *dp = spa_get_dsl(spa);
uint64_t min_bytes = zfs_dirty_data_max *
zfs_vdev_async_write_active_min_dirty_percent / 100;
uint64_t max_bytes = zfs_dirty_data_max *
zfs_vdev_async_write_active_max_dirty_percent / 100;
/*
* Async writes may occur before the assignment of the spa's
* dsl_pool_t if a self-healing zio is issued prior to the
* completion of dmu_objset_open_impl().
*/
if (dp == NULL)
return (zfs_vdev_async_write_max_active);
/*
* Sync tasks correspond to interactive user actions. To reduce the
* execution time of those actions we push data out as fast as possible.
*/
if (spa_has_pending_synctask(spa))
return (zfs_vdev_async_write_max_active);
dirty = dp->dp_dirty_total;
if (dirty < min_bytes)
return (zfs_vdev_async_write_min_active);
if (dirty > max_bytes)
return (zfs_vdev_async_write_max_active);
/*
* linear interpolation:
* slope = (max_writes - min_writes) / (max_bytes - min_bytes)
* move right by min_bytes
* move up by min_writes
*/
writes = (dirty - min_bytes) *
(zfs_vdev_async_write_max_active -
zfs_vdev_async_write_min_active) /
(max_bytes - min_bytes) +
zfs_vdev_async_write_min_active;
ASSERT3U(writes, >=, zfs_vdev_async_write_min_active);
ASSERT3U(writes, <=, zfs_vdev_async_write_max_active);
return (writes);
}
static void
log_internal(history_internal_events_t event, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
history_arg_t *ha;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
*/
if (tx->tx_txg == TXG_INITIAL)
return;
ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
#ifdef __DARWIN__
/*
* vsnprintf(NULL, 0, ...) is broken on darwin!
*/
{
char *buf = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP);
(void) vsnprintf(buf, HIS_MAX_RECORD_LEN, fmt, adx);
ha->ha_history_str = strdup(buf);
kmem_free(buf, HIS_MAX_RECORD_LEN);
}
#else
ha->ha_history_str = kmem_alloc(vsnprintf(NULL, 0, fmt, adx) + 1,
KM_SLEEP);
(void) vsprintf(ha->ha_history_str, fmt, adx);
#endif /* __DARWIN__ */
ha->ha_log_type = LOG_INTERNAL;
ha->ha_event = event;
ha->ha_zone = NULL;
ha->ha_uid = 0;
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(spa, ha, tx);
} else {
dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
spa_history_log_sync, spa, ha, 0, tx);
}
/* spa_history_log_sync() will free ha and strings */
}
zcrypt_keystore_node_t *
zcrypt_keystore_find_node(spa_t *spa, uint64_t dsobj,
boolean_t config_rwlock_held)
{
zcrypt_keystore_node_t search;
zcrypt_keystore_node_t *found = NULL;
rw_enter(&spa->spa_keystore->sk_lock, RW_READER);
if (avl_is_empty(&spa->spa_keystore->sk_dslkeys))
goto out;
search.skn_os = dsobj;
found = avl_find(&spa->spa_keystore->sk_dslkeys, &search, NULL);
if (found == NULL) {
int error;
dsl_pool_t *dp = spa_get_dsl(spa);
dsl_dataset_t *ds;
boolean_t need_lock;
rw_exit(&spa->spa_keystore->sk_lock);
need_lock = !dsl_pool_sync_context(dp) && !config_rwlock_held;
if (need_lock)
rw_enter(&dp->dp_config_rwlock, RW_READER);
error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
if (need_lock)
rw_exit(&dp->dp_config_rwlock);
rw_enter(&spa->spa_keystore->sk_lock, RW_READER);
if (!error) {
if (dsl_dataset_is_snapshot(ds)) {
search.skn_os =
ds->ds_dir->dd_phys->dd_head_dataset_obj;
found = avl_find(&spa->spa_keystore->sk_dslkeys,
&search, NULL);
}
dsl_dataset_rele(ds, FTAG);
}
}
out:
rw_exit(&spa->spa_keystore->sk_lock);
return (found);
}
static void
traverse_zil(traverse_data_t *td, zil_header_t *zh)
{
uint64_t claim_txg = zh->zh_claim_txg;
zilog_t *zilog;
/*
* We only want to visit blocks that have been claimed but not yet
* replayed; plus, in read-only mode, blocks that are already stable.
*/
if (claim_txg == 0 && spa_writeable(td->td_spa))
return;
zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
(void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
claim_txg);
zil_free(zilog);
}
void
spa_feature_create_zap_objects(spa_t *spa, dmu_tx_t *tx)
{
/*
* We create feature flags ZAP objects in two instances: during pool
* creation and during pool upgrade.
*/
ASSERT(dsl_pool_sync_context(spa_get_dsl(spa)) || (!spa->spa_sync_on &&
tx->tx_txg == TXG_INITIAL));
spa->spa_feat_for_read_obj = zap_create_link(spa->spa_meta_objset,
DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FEATURES_FOR_READ, tx);
spa->spa_feat_for_write_obj = zap_create_link(spa->spa_meta_objset,
DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FEATURES_FOR_WRITE, tx);
spa->spa_feat_desc_obj = zap_create_link(spa->spa_meta_objset,
DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FEATURE_DESCRIPTIONS, tx);
}
/*
* The nvlist will be consumed by this call.
*/
static void
log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
char *msg;
va_list adx1;
int size;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
* Likewise if the pool is not writeable.
*/
if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
fnvlist_free(nvl);
return;
}
va_copy(adx1, adx);
size = vsnprintf(NULL, 0, fmt, adx1) + 1;
msg = kmem_alloc(size, KM_PUSHPAGE);
va_end(adx1);
va_copy(adx1, adx);
(void) vsprintf(msg, fmt, adx1);
va_end(adx1);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
kmem_free(msg, size);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(nvl, tx);
} else {
dsl_sync_task_nowait(spa_get_dsl(spa),
spa_history_log_sync, nvl, 0, tx);
}
/* spa_history_log_sync() will free nvl */
}
void
spa_history_log_internal(history_internal_events_t event, spa_t *spa,
dmu_tx_t *tx, const char *fmt, ...)
{
dmu_tx_t *htx = tx;
va_list adx;
/* create a tx if we didn't get one */
if (tx == NULL) {
htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
dmu_tx_abort(htx);
return;
}
}
va_start(adx, fmt);
log_internal(event, spa, htx, fmt, adx);
va_end(adx);
/* if we didn't get a tx from the caller, commit the one we made */
if (tx == NULL)
dmu_tx_commit(htx);
}
static void
zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
{
zil_replay_arg_t *zr = zra;
const zil_header_t *zh = zilog->zl_header;
uint64_t reclen = lr->lrc_reclen;
uint64_t txtype = lr->lrc_txtype;
char *name;
int pass, error;
if (!zilog->zl_replay) /* giving up */
return;
if (lr->lrc_txg < claim_txg) /* already committed */
return;
if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */
return;
/* Strip case-insensitive bit, still present in log record */
txtype &= ~TX_CI;
if (txtype == 0 || txtype >= TX_MAX_TYPE) {
error = EINVAL;
goto bad;
}
/*
* Make a copy of the data so we can revise and extend it.
*/
bcopy(lr, zr->zr_lrbuf, reclen);
/*
* The log block containing this lr may have been byteswapped
* so that we can easily examine common fields like lrc_txtype.
* However, the log is a mix of different data types, and only the
* replay vectors know how to byteswap their records. Therefore, if
* the lr was byteswapped, undo it before invoking the replay vector.
*/
if (zr->zr_byteswap)
byteswap_uint64_array(zr->zr_lrbuf, reclen);
/*
* We must now do two things atomically: replay this log record,
* and update the log header sequence number to reflect the fact that
* we did so. At the end of each replay function the sequence number
* is updated if we are in replay mode.
*/
for (pass = 1; pass <= 2; pass++) {
zilog->zl_replaying_seq = lr->lrc_seq;
/* Only byteswap (if needed) on the 1st pass. */
error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lrbuf,
zr->zr_byteswap && pass == 1);
if (!error)
return;
/*
* The DMU's dnode layer doesn't see removes until the txg
* commits, so a subsequent claim can spuriously fail with
* EEXIST. So if we receive any error we try syncing out
* any removes then retry the transaction.
*/
if (pass == 1)
txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0);
}
bad:
ASSERT(error);
name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
dmu_objset_name(zr->zr_os, name);
cmn_err(CE_WARN, "ZFS replay transaction error %d, "
"dataset %s, seq 0x%llx, txtype %llu %s\n",
error, name, (u_longlong_t)lr->lrc_seq, (u_longlong_t)txtype,
(lr->lrc_txtype & TX_CI) ? "CI" : "");
zilog->zl_replay = B_FALSE;
kmem_free(name, MAXNAMELEN);
}
/*
* Read out the command history.
*/
int
spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
{
objset_t *mos = spa->spa_meta_objset;
dmu_buf_t *dbp;
uint64_t read_len, phys_read_off, phys_eof;
uint64_t leftover = 0;
spa_history_phys_t *shpp;
int err;
/*
* If the command history doesn't exist (older pool),
* that's ok, just return ENOENT.
*/
if (!spa->spa_history)
return (ENOENT);
/*
* The history is logged asynchronously, so when they request
* the first chunk of history, make sure everything has been
* synced to disk so that we get it.
*/
if (*offp == 0 && spa_writeable(spa))
txg_wait_synced(spa_get_dsl(spa), 0);
if ((err = dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp)) != 0)
return (err);
shpp = dbp->db_data;
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbp, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_SPA_HISTORY_OFFSETS);
}
#endif
mutex_enter(&spa->spa_history_lock);
phys_eof = spa_history_log_to_phys(shpp->sh_eof, shpp);
if (*offp < shpp->sh_pool_create_len) {
/* read in just the zpool create history */
phys_read_off = *offp;
read_len = MIN(*len, shpp->sh_pool_create_len -
phys_read_off);
} else {
/*
* Need to reset passed in offset to BOF if the passed in
* offset has since been overwritten.
*/
*offp = MAX(*offp, shpp->sh_bof);
phys_read_off = spa_history_log_to_phys(*offp, shpp);
/*
* Read up to the minimum of what the user passed down or
* the EOF (physical or logical). If we hit physical EOF,
* use 'leftover' to read from the physical BOF.
*/
if (phys_read_off <= phys_eof) {
read_len = MIN(*len, phys_eof - phys_read_off);
} else {
read_len = MIN(*len,
shpp->sh_phys_max_off - phys_read_off);
if (phys_read_off + *len > shpp->sh_phys_max_off) {
leftover = MIN(*len - read_len,
phys_eof - shpp->sh_pool_create_len);
}
}
}
/* offset for consumer to use next */
*offp += read_len + leftover;
/* tell the consumer how much you actually read */
*len = read_len + leftover;
if (read_len == 0) {
mutex_exit(&spa->spa_history_lock);
dmu_buf_rele(dbp, FTAG);
return (0);
}
err = dmu_read(mos, spa->spa_history, phys_read_off, read_len, buf,
DMU_READ_PREFETCH);
if (leftover && err == 0) {
err = dmu_read(mos, spa->spa_history, shpp->sh_pool_create_len,
leftover, buf + read_len, DMU_READ_PREFETCH);
}
mutex_exit(&spa->spa_history_lock);
dmu_buf_rele(dbp, FTAG);
return (err);
}
static void
zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
{
zil_replay_arg_t *zr = zra;
const zil_header_t *zh = zilog->zl_header;
uint64_t reclen = lr->lrc_reclen;
uint64_t txtype = lr->lrc_txtype;
char *name;
int pass, error;
if (!zilog->zl_replay) /* giving up */
return;
if (lr->lrc_txg < claim_txg) /* already committed */
return;
if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */
return;
/* Strip case-insensitive bit, still present in log record */
txtype &= ~TX_CI;
if (txtype == 0 || txtype >= TX_MAX_TYPE) {
error = EINVAL;
goto bad;
}
/*
* Make a copy of the data so we can revise and extend it.
*/
bcopy(lr, zr->zr_lrbuf, reclen);
/*
* The log block containing this lr may have been byteswapped
* so that we can easily examine common fields like lrc_txtype.
* However, the log is a mix of different data types, and only the
* replay vectors know how to byteswap their records. Therefore, if
* the lr was byteswapped, undo it before invoking the replay vector.
*/
if (zr->zr_byteswap)
byteswap_uint64_array(zr->zr_lrbuf, reclen);
/*
* If this is a TX_WRITE with a blkptr, suck in the data.
*/
if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) {
lr_write_t *lrw = (lr_write_t *)lr;
blkptr_t *wbp = &lrw->lr_blkptr;
uint64_t wlen = lrw->lr_length;
char *wbuf = zr->zr_lrbuf + reclen;
if (BP_IS_HOLE(wbp)) { /* compressed to a hole */
bzero(wbuf, wlen);
} else {
/*
* A subsequent write may have overwritten this block,
* in which case wbp may have been been freed and
* reallocated, and our read of wbp may fail with a
* checksum error. We can safely ignore this because
* the later write will provide the correct data.
*/
zbookmark_t zb;
zb.zb_objset = dmu_objset_id(zilog->zl_os);
zb.zb_object = lrw->lr_foid;
zb.zb_level = -1;
zb.zb_blkid = lrw->lr_offset / BP_GET_LSIZE(wbp);
(void) zio_wait(zio_read(NULL, zilog->zl_spa,
wbp, wbuf, BP_GET_LSIZE(wbp), NULL, NULL,
ZIO_PRIORITY_SYNC_READ,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &zb));
(void) memmove(wbuf, wbuf + lrw->lr_blkoff, wlen);
}
}
/*
* We must now do two things atomically: replay this log record,
* and update the log header sequence number to reflect the fact that
* we did so. At the end of each replay function the sequence number
* is updated if we are in replay mode.
*/
for (pass = 1; pass <= 2; pass++) {
zilog->zl_replaying_seq = lr->lrc_seq;
/* Only byteswap (if needed) on the 1st pass. */
error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lrbuf,
zr->zr_byteswap && pass == 1);
if (!error)
return;
/*
* The DMU's dnode layer doesn't see removes until the txg
* commits, so a subsequent claim can spuriously fail with
* EEXIST. So if we receive any error we try syncing out
* any removes then retry the transaction.
*/
if (pass == 1)
txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0);
}
bad:
ASSERT(error);
name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
dmu_objset_name(zr->zr_os, name);
cmn_err(CE_WARN, "ZFS replay transaction error %d, "
"dataset %s, seq 0x%llx, txtype %llu %s\n",
error, name, (u_longlong_t)lr->lrc_seq, (u_longlong_t)txtype,
(lr->lrc_txtype & TX_CI) ? "CI" : "");
zilog->zl_replay = B_FALSE;
kmem_free(name, MAXNAMELEN);
}
/*
* same as dsl_open_dir, ignore the first component of name and use the
* spa instead
*/
int
dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
dsl_dir_t **ddp, const char **tailp)
{
char buf[MAXNAMELEN];
const char *next, *nextnext = NULL;
int err;
dsl_dir_t *dd;
dsl_pool_t *dp;
uint64_t ddobj;
int openedspa = FALSE;
dprintf("%s\n", name);
err = getcomponent(name, buf, &next);
if (err)
return (err);
if (spa == NULL) {
err = spa_open(buf, &spa, FTAG);
if (err) {
dprintf("spa_open(%s) failed\n", buf);
return (err);
}
openedspa = TRUE;
/* XXX this assertion belongs in spa_open */
ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
}
dp = spa_get_dsl(spa);
rw_enter(&dp->dp_config_rwlock, RW_READER);
err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
if (err) {
rw_exit(&dp->dp_config_rwlock);
if (openedspa)
spa_close(spa, FTAG);
return (err);
}
while (next != NULL) {
dsl_dir_t *child_ds;
err = getcomponent(next, buf, &nextnext);
if (err)
break;
ASSERT(next[0] != '\0');
if (next[0] == '@')
break;
dprintf("looking up %s in obj%lld\n",
buf, dd->dd_phys->dd_child_dir_zapobj);
err = zap_lookup(dp->dp_meta_objset,
dd->dd_phys->dd_child_dir_zapobj,
buf, sizeof (ddobj), 1, &ddobj);
if (err) {
if (err == ENOENT)
err = 0;
break;
}
err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
if (err)
break;
dsl_dir_close(dd, tag);
dd = child_ds;
next = nextnext;
}
rw_exit(&dp->dp_config_rwlock);
if (err) {
dsl_dir_close(dd, tag);
if (openedspa)
spa_close(spa, FTAG);
return (err);
}
/*
* It's an error if there's more than one component left, or
* tailp==NULL and there's any component left.
*/
if (next != NULL &&
(tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
/* bad path name */
dsl_dir_close(dd, tag);
dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
err = ENOENT;
}
if (tailp)
*tailp = next;
if (openedspa)
spa_close(spa, FTAG);
*ddp = dd;
return (err);
}
