bool sockinfo::detach_receiver(flow_tuple_with_local_if &flow_key)
{
si_logdbg("Unregistering receiver: %s", flow_key.to_str());
// TODO ALEXR: DO we need to return a 3 tuple instead of a 5 tuple being removed?
// if (peer_ip != INADDR_ANY && peer_port != INPORT_ANY);
// Find ring associated with this tuple
rx_flow_map_t::iterator rx_flow_iter = m_rx_flow_map.find(flow_key);
BULLSEYE_EXCLUDE_BLOCK_START
if (rx_flow_iter == m_rx_flow_map.end()) {
si_logdbg("Failed to find ring associated with: %s", flow_key.to_str());
return false;
}
BULLSEYE_EXCLUDE_BLOCK_END
ring* p_ring = rx_flow_iter->second;
si_logdbg("Detaching %s from ring %p", flow_key.to_str(), p_ring);
// Detach tuple
unlock_rx_q();
p_ring->detach_flow(flow_key, this);
lock_rx_q();
// Un-map flow from local map
#ifndef DEFINED_SOCKETXTREME // is not defined
rx_del_ring_cb(flow_key, p_ring);
#endif // DEFINED_SOCKETXTREME
m_rx_flow_map.erase(rx_flow_iter);
return destroy_nd_resources((const ip_address)flow_key.get_local_if());
}
int sockinfo::ioctl(unsigned long int __request, unsigned long int __arg) throw (vma_error)
{
int *p_arg = (int *)__arg;
switch (__request) {
case FIONBIO:
{
si_logdbg("request=FIONBIO, arg=%d", *p_arg);
if (*p_arg)
set_blocking(false);
else
set_blocking(true);
}
break;
default:
char buf[128];
snprintf(buf, sizeof(buf), "unimplemented ioctl request=%#x, flags=%#x", (unsigned)__request, (unsigned)__arg);
buf[ sizeof(buf)-1 ] = '\0';
VLOG_PRINTF_INFO(safe_mce_sys().exception_handling.get_log_severity(), "%s", buf);
int rc = handle_exception_flow();
switch (rc) {
case -1:
return rc;
case -2:
vma_throw_object_with_msg(vma_unsupported_api, buf);
}
break;
}
si_logdbg("going to OS for ioctl request=%d, flags=%x", __request, __arg);
return orig_os_api.ioctl(m_fd, __request, __arg);
}
net_device_resources_t* sockinfo::create_nd_resources(const ip_address ip_local)
{
net_device_resources_t* p_nd_resources = NULL;
// Check if we are already registered to net_device with the local ip as observers
rx_net_device_map_t::iterator rx_nd_iter = m_rx_nd_map.find(ip_local.get_in_addr());
if (rx_nd_iter == m_rx_nd_map.end()) {
// Need to register as observer to net_device
net_device_resources_t nd_resources;
nd_resources.refcnt = 0;
nd_resources.p_nde = NULL;
nd_resources.p_ndv = NULL;
nd_resources.p_ring = NULL;
BULLSEYE_EXCLUDE_BLOCK_START
cache_entry_subject<ip_address, net_device_val*>* p_ces = NULL;
if (!g_p_net_device_table_mgr->register_observer(ip_local, &m_rx_nd_observer, &p_ces)) {
si_logdbg("Failed registering as observer for local ip %s", ip_local.to_str().c_str());
goto err;
}
nd_resources.p_nde = (net_device_entry*)p_ces;
if (!nd_resources.p_nde) {
si_logerr("Got NULL net_devide_entry for local ip %s", ip_local.to_str().c_str());
goto err;
}
if (!nd_resources.p_nde->get_val(nd_resources.p_ndv)) {
si_logerr("Got net_device_val=NULL (interface is not offloaded) for local ip %s", ip_local.to_str().c_str());
goto err;
}
unlock_rx_q();
m_rx_ring_map_lock.lock();
resource_allocation_key *key;
if (m_rx_ring_map.size() && m_ring_alloc_logic.is_logic_support_migration()) {
key = m_ring_alloc_logic.get_key();
} else {
key = m_ring_alloc_logic.create_new_key(ip_local.get_in_addr());
}
nd_resources.p_ring = nd_resources.p_ndv->reserve_ring(key);
m_rx_ring_map_lock.unlock();
lock_rx_q();
if (!nd_resources.p_ring) {
si_logdbg("Failed to reserve ring for allocation key %s on ip %s",
m_ring_alloc_logic.get_key()->to_str(), ip_local.to_str().c_str());
goto err;
}
// Add new net_device to rx_map
m_rx_nd_map[ip_local.get_in_addr()] = nd_resources;
rx_nd_iter = m_rx_nd_map.find(ip_local.get_in_addr());
if (rx_nd_iter == m_rx_nd_map.end()) {
si_logerr("Failed to find rx_nd_iter");
goto err;
}
BULLSEYE_EXCLUDE_BLOCK_END
}
void sockinfo::set_blocking(bool is_blocked)
{
if (is_blocked) {
si_logdbg("set socket to blocked mode");
m_b_blocking = true;
}
else {
si_logdbg("set socket to non-blocking mode");
m_b_blocking = false;
}
// Update statistics info
m_p_socket_stats->b_blocking = m_b_blocking;
}
int sockinfo::getsockopt(int __level, int __optname, void *__optval, socklen_t *__optlen)
{
int ret = -1;
switch (__level) {
case SOL_SOCKET:
switch(__optname) {
case SO_VMA_USER_DATA:
if (*__optlen == sizeof(m_fd_context)) {
*(void **)__optval = m_fd_context;
ret = 0;
} else {
errno = EINVAL;
}
break;
case SO_VMA_FLOW_TAG:
if (*__optlen >= sizeof(uint32_t)) {
*(uint32_t*)__optval = m_flow_tag_id;
ret = 0;
} else {
errno = EINVAL;
}
break;
case SO_MAX_PACING_RATE:
if (*__optlen == sizeof(struct vma_rate_limit_t)) {
*(struct vma_rate_limit_t*)__optval = m_so_ratelimit;
*__optlen = sizeof(struct vma_rate_limit_t);
si_logdbg("(SO_MAX_PACING_RATE) value: %d, %d, %d",
(*(struct vma_rate_limit_t*)__optval).rate,
(*(struct vma_rate_limit_t*)__optval).max_burst_sz,
(*(struct vma_rate_limit_t*)__optval).typical_pkt_sz);
} else if (*__optlen == sizeof(uint32_t)) {
*(uint32_t*)__optval = KB_TO_BYTE(m_so_ratelimit.rate);
*__optlen = sizeof(uint32_t);
si_logdbg("(SO_MAX_PACING_RATE) value: %d",
*(int *)__optval);
ret = 0;
} else {
errno = EINVAL;
}
break;
}
}
return ret;
}
int sockinfo::fcntl(int __cmd, unsigned long int __arg)
{
switch (__cmd) {
case F_SETFL:
{
si_logdbg("cmd=F_SETFL, arg=%#x", __arg);
if (__arg & O_NONBLOCK)
set_blocking(false);
else
set_blocking(true);
}
break;
case F_GETFL: /* Get file status flags. */
si_logfunc("cmd=F_GETFL, arg=%#x", __arg);
break;
case F_GETFD: /* Get file descriptor flags. */
si_logfunc("cmd=F_GETFD, arg=%#x", __arg);
break;
case F_SETFD: /* Set file descriptor flags. */
si_logfunc("cmd=F_SETFD, arg=%#x", __arg);
break;
default:
char buf[128];
snprintf(buf, sizeof(buf), "unimplemented fcntl cmd=%#x, arg=%#x", (unsigned)__cmd, (unsigned)__arg);
buf[ sizeof(buf)-1 ] = '\0';
VLOG_PRINTF_INFO(safe_mce_sys().exception_handling.get_log_severity(), "%s", buf);
int rc = handle_exception_flow();
switch (rc) {
case -1:
return rc;
case -2:
vma_throw_object_with_msg(vma_unsupported_api, buf);
}
break;
}
si_logdbg("going to OS for fcntl cmd=%d, arg=%#x", __cmd, __arg);
return orig_os_api.fcntl(m_fd, __cmd, __arg);
}
bool sockinfo::attach_receiver(flow_tuple_with_local_if &flow_key)
{
// This function should be called from within mutex protected context of the sockinfo!!!
si_logdbg("Attaching to %s", flow_key.to_str());
// Protect against local loopback used as local_if & peer_ip
// rdma_cm will accept it but we don't want to offload it
if (flow_key.is_local_loopback()) {
si_logdbg("VMA does not offload local loopback IP address");
return false;
}
if (m_rx_flow_map.find(flow_key) != m_rx_flow_map.end()) {
si_logdbg("already attached %s", flow_key.to_str());
return false;
}
// Allocate resources on specific interface (create ring)
net_device_resources_t* p_nd_resources = create_nd_resources((const ip_address)flow_key.get_local_if());
if (NULL == p_nd_resources) {
// any error which occurred inside create_nd_resources() was already printed. No need to reprint errors here
return false;
}
// Map flow in local map
m_rx_flow_map[flow_key] = p_nd_resources->p_ring;
#ifndef DEFINED_SOCKETXTREME // is not defined
// Save the new CQ from ring
rx_add_ring_cb(flow_key, p_nd_resources->p_ring);
#endif // DEFINED_SOCKETXTREME
// Attach tuple
BULLSEYE_EXCLUDE_BLOCK_START
unlock_rx_q();
if (!p_nd_resources->p_ring->attach_flow(flow_key, this)) {
lock_rx_q();
si_logdbg("Failed to attach %s to ring %p", flow_key.to_str(), p_nd_resources->p_ring);
return false;
}
set_rx_packet_processor();
lock_rx_q();
BULLSEYE_EXCLUDE_BLOCK_END
// Registered as receiver successfully
si_logdbg("Attached %s to ring %p", flow_key.to_str(), p_nd_resources->p_ring);
// Verify 5 tuple over 3 tuple
if (flow_key.is_5_tuple())
{
// Check and remove lesser 3 tuple
flow_tuple_with_local_if flow_key_3t(flow_key.get_dst_ip(), flow_key.get_dst_port(), INADDR_ANY, INPORT_ANY, flow_key.get_protocol(), flow_key.get_local_if());
rx_flow_map_t::iterator rx_flow_iter = m_rx_flow_map.find(flow_key_3t);
if (rx_flow_iter != m_rx_flow_map.end()) {
si_logdbg("Removing (and detaching) 3 tuple now that we added a stronger 5 tuple");
detach_receiver(flow_key_3t);
}
}
return true;
}
int sockinfo::setsockopt(int __level, int __optname, const void *__optval, socklen_t __optlen)
{
int ret = SOCKOPT_PASS_TO_OS;
if (__level == SOL_SOCKET) {
switch(__optname) {
case SO_VMA_USER_DATA:
if (__optlen == sizeof(m_fd_context)) {
m_fd_context = *(void **)__optval;
ret = SOCKOPT_INTERNAL_VMA_SUPPORT;
} else {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EINVAL;
}
break;
case SO_VMA_RING_USER_MEMORY:
if (__optval) {
if (__optlen == sizeof(iovec)) {
iovec *attr = (iovec *)__optval;
m_ring_alloc_log_rx.set_memory_descriptor(*attr);
m_ring_alloc_logic = ring_allocation_logic_rx(get_fd(), m_ring_alloc_log_rx, this);
if (m_p_rx_ring || m_rx_ring_map.size()) {
si_logwarn("user asked to assign memory for "
"RX ring but ring already exists");
}
ret = SOCKOPT_INTERNAL_VMA_SUPPORT;
} else {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EINVAL;
si_logdbg("SOL_SOCKET, SO_VMA_RING_USER_MEMORY - "
"bad length expected %d got %d",
sizeof(iovec), __optlen);
}
}
else {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EINVAL;
si_logdbg("SOL_SOCKET, SO_VMA_RING_USER_MEMORY - NOT HANDLED, optval == NULL");
}
break;
case SO_VMA_FLOW_TAG:
if (__optval) {
if (__optlen == sizeof(uint32_t)) {
if (set_flow_tag(*(uint32_t*)__optval)) {
si_logdbg("SO_VMA_FLOW_TAG, set "
"socket %s to flow id %d",
m_fd, m_flow_tag_id);
// not supported in OS
ret = SOCKOPT_INTERNAL_VMA_SUPPORT;
} else {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EINVAL;
}
} else {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EINVAL;
si_logdbg("SO_VMA_FLOW_TAG, bad length "
"expected %d got %d",
sizeof(uint32_t), __optlen);
break;
}
} else {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EINVAL;
si_logdbg("SO_VMA_FLOW_TAG - NOT HANDLED, "
"optval == NULL");
}
break;
case SO_TIMESTAMP:
case SO_TIMESTAMPNS:
if (__optval) {
m_b_rcvtstamp = *(bool*)__optval;
if (__optname == SO_TIMESTAMPNS)
m_b_rcvtstampns = m_b_rcvtstamp;
si_logdbg("SOL_SOCKET, %s=%s", setsockopt_so_opt_to_str(__optname), (m_b_rcvtstamp ? "true" : "false"));
}
else {
si_logdbg("SOL_SOCKET, %s=\"???\" - NOT HANDLED, optval == NULL", setsockopt_so_opt_to_str(__optname));
}
break;
case SO_TIMESTAMPING:
if (__optval) {
uint8_t val = *(uint8_t*)__optval;
// SOF_TIMESTAMPING_TX_SOFTWARE and SOF_TIMESTAMPING_TX_HARDWARE is NOT supported.
if (val & (SOF_TIMESTAMPING_TX_SOFTWARE | SOF_TIMESTAMPING_TX_HARDWARE)) {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EOPNOTSUPP;
si_logdbg("SOL_SOCKET, SOF_TIMESTAMPING_TX_SOFTWARE and SOF_TIMESTAMPING_TX_HARDWARE is not supported, errno set to EOPNOTSUPP");
}
if (val & (SOF_TIMESTAMPING_RAW_HARDWARE | SOF_TIMESTAMPING_RX_HARDWARE)) {
if (g_p_ib_ctx_handler_collection->get_ctx_time_conversion_mode() == TS_CONVERSION_MODE_DISABLE){
if (safe_mce_sys().hw_ts_conversion_mode == TS_CONVERSION_MODE_DISABLE) {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EPERM;
si_logdbg("SOL_SOCKET, SOF_TIMESTAMPING_RAW_HARDWARE and SOF_TIMESTAMPING_RX_HARDWARE socket options were disabled (VMA_HW_TS_CONVERSION = %d) , errno set to EPERM", TS_CONVERSION_MODE_DISABLE);
} else {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = ENODEV;
si_logdbg("SOL_SOCKET, SOF_TIMESTAMPING_RAW_HARDWARE and SOF_TIMESTAMPING_RX_HARDWARE is not supported by device(s), errno set to ENODEV");
}
}
}
m_n_tsing_flags = val;
si_logdbg("SOL_SOCKET, SO_TIMESTAMPING=%u", m_n_tsing_flags);
}
else {
si_logdbg("SOL_SOCKET, %s=\"???\" - NOT HANDLED, optval == NULL", setsockopt_so_opt_to_str(__optname));
}
break;
default:
break;
}
} else if (__level == IPPROTO_IP) {
switch(__optname) {
case IP_TTL:
if (__optlen < sizeof(m_n_uc_ttl)) {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EINVAL;
} else {
int val = __optlen < sizeof(val) ? (uint8_t) *(uint8_t *)__optval : (int) *(int *)__optval;
if (val != -1 && (val < 1 || val > 255)) {
ret = SOCKOPT_NO_VMA_SUPPORT;
errno = EINVAL;
} else {
m_n_uc_ttl = (val == -1) ? safe_mce_sys().sysctl_reader.get_net_ipv4_ttl() : (uint8_t) val;
header_ttl_updater du(m_n_uc_ttl, false);
update_header_field(&du);
si_logdbg("IPPROTO_IP, optname=IP_TTL (%d)", m_n_uc_ttl);
}
}
break;
default:
break;
}
}
si_logdbg("ret (%d)", ret);
return ret;
}
bool sockinfo::attach_receiver(flow_tuple_with_local_if &flow_key)
{
// This function should be called from within mutex protected context of the sockinfo!!!
si_logdbg("Attaching to %s", flow_key.to_str());
// Protect against local loopback used as local_if & peer_ip
// rdma_cm will accept it but we don't want to offload it
if (flow_key.is_local_loopback()) {
si_logdbg("VMA does not offload local loopback IP address");
return false;
}
if (m_rx_flow_map.find(flow_key) != m_rx_flow_map.end()) {
si_logdbg("already attached %s", flow_key.to_str());
return false;
}
net_device_resources_t* p_nd_resources = NULL;
// Check if we are already registered to net_device with the local ip as observers
ip_address ip_local(flow_key.get_local_if());
rx_net_device_map_t::iterator rx_nd_iter = m_rx_nd_map.find(ip_local.get_in_addr());
if (rx_nd_iter == m_rx_nd_map.end()) {
// Need to register as observer to net_device
net_device_resources_t nd_resources;
nd_resources.refcnt = 0;
nd_resources.p_nde = NULL;
nd_resources.p_ndv = NULL;
nd_resources.p_ring = NULL;
BULLSEYE_EXCLUDE_BLOCK_START
cache_entry_subject<ip_address, net_device_val*>* p_ces = NULL;
if (!g_p_net_device_table_mgr->register_observer(ip_local, &m_rx_nd_observer, &p_ces)) {
si_logdbg("Failed registering as observer for local ip %s", ip_local.to_str().c_str());
return false;
}
nd_resources.p_nde = (net_device_entry*)p_ces;
if (!nd_resources.p_nde) {
si_logerr("Got NULL net_devide_entry for local ip %s", ip_local.to_str().c_str());
return false;
}
if (!nd_resources.p_nde->get_val(nd_resources.p_ndv)) {
si_logerr("Got net_device_val=NULL (interface is not offloaded) for local ip %s", ip_local.to_str().c_str());
return false;
}
unlock_rx_q();
m_rx_ring_map_lock.lock();
resource_allocation_key key = 0;
if (m_rx_ring_map.size()) {
key = m_ring_alloc_logic.get_key();
} else {
key = m_ring_alloc_logic.create_new_key();
}
nd_resources.p_ring = nd_resources.p_ndv->reserve_ring(key);
m_rx_ring_map_lock.unlock();
lock_rx_q();
if (!nd_resources.p_ring) {
si_logdbg("Failed to reserve ring for allocation key %d on lip %s", m_ring_alloc_logic.get_key(), ip_local.to_str().c_str());
return false;
}
// Add new net_device to rx_map
m_rx_nd_map[ip_local.get_in_addr()] = nd_resources;
rx_nd_iter = m_rx_nd_map.find(ip_local.get_in_addr());
if (rx_nd_iter == m_rx_nd_map.end()) {
si_logerr("Failed to find rx_nd_iter");
return false;
}
BULLSEYE_EXCLUDE_BLOCK_END
}