int crypto_run(struct fcrypt *fcr, struct kernel_crypt_op *kcop)
{
struct csession *ses_ptr;
struct crypt_op *cop = &kcop->cop;
int ret;
if (unlikely(cop->op != COP_ENCRYPT && cop->op != COP_DECRYPT)) {
ddebug(1, "invalid operation op=%u", cop->op);
return -EINVAL;
}
/* this also enters ses_ptr->sem */
ses_ptr = crypto_get_session_by_sid(fcr, cop->ses);
if (unlikely(!ses_ptr)) {
derr(1, "invalid session ID=0x%08X", cop->ses);
return -EINVAL;
}
if (ses_ptr->hdata.init != 0 && (cop->flags == 0 || cop->flags & COP_FLAG_RESET)) {
ret = cryptodev_hash_reset(&ses_ptr->hdata);
if (unlikely(ret)) {
derr(1, "error in cryptodev_hash_reset()");
goto out_unlock;
}
}
if (ses_ptr->cdata.init != 0) {
int blocksize = ses_ptr->cdata.blocksize;
if (unlikely(cop->len % blocksize)) {
derr(1, "data size (%u) isn't a multiple of block size (%u)",
cop->len, blocksize);
ret = -EINVAL;
goto out_unlock;
}
cryptodev_cipher_set_iv(&ses_ptr->cdata, kcop->iv,
min(ses_ptr->cdata.ivsize, kcop->ivlen));
}
if (likely(cop->len)) {
if (cop->flags & COP_FLAG_NO_ZC) {
if (unlikely(ses_ptr->alignmask && !IS_ALIGNED((unsigned long)cop->src, ses_ptr->alignmask))) {
dwarning(2, "source address %p is not %d byte aligned - disabling zero copy",
cop->src, ses_ptr->alignmask + 1);
cop->flags &= ~COP_FLAG_NO_ZC;
}
if (unlikely(ses_ptr->alignmask && !IS_ALIGNED((unsigned long)cop->dst, ses_ptr->alignmask))) {
dwarning(2, "destination address %p is not %d byte aligned - disabling zero copy",
cop->dst, ses_ptr->alignmask + 1);
cop->flags &= ~COP_FLAG_NO_ZC;
}
}
if (cop->flags & COP_FLAG_NO_ZC)
ret = __crypto_run_std(ses_ptr, &kcop->cop);
else
ret = __crypto_run_zc(ses_ptr, kcop);
if (unlikely(ret))
goto out_unlock;
}
if (ses_ptr->cdata.init != 0) {
cryptodev_cipher_get_iv(&ses_ptr->cdata, kcop->iv,
min(ses_ptr->cdata.ivsize, kcop->ivlen));
}
if (ses_ptr->hdata.init != 0 &&
((cop->flags & COP_FLAG_FINAL) ||
(!(cop->flags & COP_FLAG_UPDATE) || cop->len == 0))) {
ret = cryptodev_hash_final(&ses_ptr->hdata, kcop->hash_output);
if (unlikely(ret)) {
derr(0, "CryptoAPI failure: %d", ret);
goto out_unlock;
}
kcop->digestsize = ses_ptr->hdata.digestsize;
}
if (ses_ptr->rdata.init != 0 && cop->len > 0) {
kcop->rng_output = kmalloc(cop->len, GFP_KERNEL);
if (unlikely(!kcop->rng_output)) {
derr(0, "Not enough space to store %d random bytes.", cop->len);
ret = -ENOMEM;
goto out_unlock;
}
ret = cryptodev_rng_get_bytes(&ses_ptr->rdata, kcop->rng_output, cop->len);
// some RNGs return 0 for success, while
// some return the number of bytes generated
if (unlikely(ret != 0 && ret != cop->len)) {
derr(0, "RNG failure: %d", ret);
kfree(kcop->rng_output); kcop->rng_output = NULL;
goto out_unlock;
}
ret = 0;
kcop->rnglen = cop->len;
}
out_unlock:
crypto_put_session(ses_ptr);
return ret;
}
error_code meta_service::start()
{
dassert(!_started, "meta service is already started");
// init server state
error_code err = _state->initialize();
if (err != ERR_OK)
{
derror("init server_state failed, err = %s", err.to_string());
return err;
}
ddebug("init server state succeed");
// we should start the FD service to response to the workers fd request
_failure_detector = new meta_server_failure_detector(_state, this);
err = _failure_detector->start(
_opts.fd_check_interval_seconds,
_opts.fd_beacon_interval_seconds,
_opts.fd_lease_seconds,
_opts.fd_grace_seconds,
false
);
if (err != ERR_OK)
{
derror("start failure_detector failed, err = %s", err.to_string());
return err;
}
// should register rpc handlers before acquiring leader lock, so that this meta service
// can tell others who is the current leader
register_rpc_handlers();
// become leader
_failure_detector->acquire_leader_lock();
dassert(_failure_detector->is_primary(), "must be primary at this point");
ddebug("hahaha, I got the primary lock! now start to recover server state");
// recover server state
while ((err = _state->on_become_leader()) != ERR_OK)
{
derror("recover server state failed, err = %s, retry ...", err.to_string());
}
// create server load balancer
// TODO: create per app server load balancer
const char* server_load_balancer = dsn_config_get_value_string(
"meta_server",
"server_load_balancer_type",
"simple_stateful_load_balancer",
"server_load_balancer provider type"
);
_balancer = dsn::utils::factory_store< ::dsn::dist::server_load_balancer>::create(
server_load_balancer,
PROVIDER_TYPE_MAIN,
_state
);
_failure_detector->sync_node_state_and_start_service();
ddebug("start meta_service succeed");
return ERR_OK;
}
void replica::on_prepare_reply(std::pair<mutation_ptr, partition_status> pr, int err, message_ptr& request, message_ptr& reply)
{
check_hashed_access();
mutation_ptr& mu = pr.first;
partition_status targetStatus = pr.second;
// skip callback for old mutations
if (mu->data.header.ballot < get_ballot() || PS_PRIMARY != status())
return;
dassert (mu->data.header.ballot == get_ballot(), "");
end_point node = request->header().to_address;
partition_status st = _primary_states.GetNodeStatus(node);
// handle reply
prepare_ack resp;
// handle error
if (err)
{
resp.err = err;
}
else
{
unmarshall(reply, resp);
ddebug(
"%s: mutation %s on_prepare_reply from %s:%d",
name(), mu->name(),
node.name.c_str(), static_cast<int>(node.port)
);
}
if (resp.err == ERR_SUCCESS)
{
dassert (resp.ballot == get_ballot(), "");
dassert (resp.decree == mu->data.header.decree, "");
switch (targetStatus)
{
case PS_SECONDARY:
dassert (_primary_states.check_exist(node, PS_SECONDARY), "");
dassert (mu->left_secondary_ack_count() > 0, "");
if (0 == mu->decrease_left_secondary_ack_count())
{
do_possible_commit_on_primary(mu);
}
break;
case PS_POTENTIAL_SECONDARY:
dassert (mu->left_potential_secondary_ack_count() > 0, "");
if (0 == mu->decrease_left_potential_secondary_ack_count())
{
do_possible_commit_on_primary(mu);
}
break;
default:
ddebug(
"%s: mutation %s prepare ack skipped coz the node is now inactive", name(), mu->name()
);
break;
}
}
// failure handling
else
{
// note targetStatus and (curent) status may diff
if (targetStatus == PS_POTENTIAL_SECONDARY)
{
dassert (mu->left_potential_secondary_ack_count() > 0, "");
if (0 == mu->decrease_left_potential_secondary_ack_count())
{
do_possible_commit_on_primary(mu);
}
}
handle_remote_failure(st, node, resp.err);
}
}
void replica::init_prepare(mutation_ptr& mu)
{
dassert (PS_PRIMARY == status(), "");
error_code err = ERR_SUCCESS;
uint8_t count = 0;
if (static_cast<int>(_primary_states.membership.secondaries.size()) + 1 < _options.mutation_2pc_min_replica_count)
{
err = ERR_NOT_ENOUGH_MEMBER;
goto ErrOut;
}
mu->data.header.last_committed_decree = last_committed_decree();
if (mu->data.header.decree == invalid_decree)
{
mu->set_id(get_ballot(), _prepare_list->max_decree() + 1);
}
else
{
mu->set_id(get_ballot(), mu->data.header.decree);
}
if (mu->data.header.decree > _prepare_list->max_decree() && _prepare_list->count() >= _options.staleness_for_commit)
{
err = ERR_CAPACITY_EXCEEDED;
goto ErrOut;
}
dassert (mu->data.header.decree > last_committed_decree(), "");
// local prepare without log
err = _prepare_list->prepare(mu, PS_PRIMARY);
if (err != ERR_SUCCESS)
{
goto ErrOut;
}
ddebug("%s: mutation %s init_prepare", name(), mu->name());
//
// TODO: bounded staleness on secondaries
//
dassert (mu->data.header.decree <= last_committed_decree() + _options.staleness_for_commit, "");
// remote prepare
dassert (mu->remote_tasks().size() == 0, "");
mu->set_left_secondary_ack_count((unsigned int)_primary_states.membership.secondaries.size());
for (auto it = _primary_states.membership.secondaries.begin(); it != _primary_states.membership.secondaries.end(); it++)
{
send_prepare_message(*it, PS_SECONDARY, mu, _options.prepare_timeout_ms_for_secondaries);
}
count = 0;
for (auto it = _primary_states.learners.begin(); it != _primary_states.learners.end(); it++)
{
if (it->second.prepare_start_decree != invalid_decree && mu->data.header.decree >= it->second.prepare_start_decree)
{
send_prepare_message(it->first, PS_POTENTIAL_SECONDARY, mu, _options.prepare_timeout_ms_for_potential_secondaries);
count++;
}
}
mu->set_left_potential_secondary_ack_count(count);
// local log
dassert (mu->data.header.log_offset == invalid_offset, "");
dassert (mu->log_task() == nullptr, "");
mu->log_task() = _stub->_log->append(mu,
LPC_WRITE_REPLICATION_LOG,
this,
std::bind(&replica::on_append_log_completed, this, mu,
std::placeholders::_1,
std::placeholders::_2),
gpid_to_hash(get_gpid())
);
if (nullptr == mu->log_task())
{
err = ERR_FILE_OPERATION_FAILED;
handle_local_failure(err);
goto ErrOut;
}
return;
ErrOut:
response_client_message(mu->client_request, err);
return;
}
void replica::on_prepare(message_ptr& request)
{
check_hashed_access();
replica_configuration rconfig;
unmarshall(request, rconfig);
mutation_ptr mu = mutation::read_from(request);
decree decree = mu->data.header.decree;
ddebug( "%s: mutation %s on_prepare", name(), mu->name());
dassert (mu->data.header.ballot == rconfig.ballot, "");
if (mu->data.header.ballot < get_ballot())
{
ddebug( "%s: mutation %s on_prepare skipped due to old view", name(), mu->name());
return;
}
// update configuration when necessary
else if (rconfig.ballot > get_ballot())
{
update_local_configuration(rconfig);
}
if (PS_INACTIVE == status() || PS_ERROR == status())
{
ddebug(
"%s: mutation %s on_prepare to %s skipped",
name(), mu->name(),
enum_to_string(status())
);
ack_prepare_message(ERR_INVALID_STATE, mu);
return;
}
else if (PS_POTENTIAL_SECONDARY == status())
{
if (_potential_secondary_states.learning_status != LearningWithPrepare && _potential_secondary_states.learning_status != LearningSucceeded)
{
ddebug(
"%s: mutation %s on_prepare to %s skipped, learnings state = %s",
name(), mu->name(),
enum_to_string(status()),
enum_to_string(_potential_secondary_states.learning_status)
);
// do not retry as there may retries later
return;
}
}
dassert (rconfig.status == status(), "");
if (decree <= last_committed_decree())
{
ack_prepare_message(ERR_SUCCESS, mu);
return;
}
// real prepare start
auto mu2 = _prepare_list->get_mutation_by_decree(decree);
if (mu2 != nullptr && mu2->data.header.ballot == mu->data.header.ballot)
{
ddebug( "%s: mutation %s redundant prepare skipped", name(), mu->name());
if (mu2->is_prepared())
{
ack_prepare_message(ERR_SUCCESS, mu);
}
return;
}
int err = _prepare_list->prepare(mu, status());
dassert (err == ERR_SUCCESS, "");
if (PS_POTENTIAL_SECONDARY == status())
{
dassert (mu->data.header.decree <= last_committed_decree() + _options.staleness_for_start_prepare_for_potential_secondary, "");
}
else
{
dassert (PS_SECONDARY == status(), "");
dassert (mu->data.header.decree <= last_committed_decree() + _options.staleness_for_commit, "");
}
// write log
dassert (mu->log_task() == nullptr, "");
mu->log_task() = _stub->_log->append(mu,
LPC_WRITE_REPLICATION_LOG,
this,
std::bind(&replica::on_append_log_completed, this, mu, std::placeholders::_1, std::placeholders::_2),
gpid_to_hash(get_gpid())
);
if (nullptr == mu->log_task())
{
err = ERR_FILE_OPERATION_FAILED;
ack_prepare_message(err, mu);
handle_local_failure(err);
}
}
