void citp_waitable_print(citp_waitable* w)
{
/* Output socket using netstat style output:
* TCP 2 0 0.0.0.0:12865 0.0.0.0:0 LISTEN
* UDP 0 0 172.16.129.131:57521 0.0.0.0:0 UDP
*/
if( CI_TCP_STATE_IS_SOCKET(w->state) ) {
ci_sock_cmn* s = CI_CONTAINER(ci_sock_cmn, b, w);
citp_waitable_obj* wo = CI_CONTAINER(citp_waitable_obj, waitable, w);
int tq = 0;
int rq = 0;
if( (w->state & CI_TCP_STATE_TCP) &&
!(w->state & CI_TCP_STATE_NOT_CONNECTED) ) {
tq = ci_tcp_sendq_n_pkts(&wo->tcp);
rq = wo->tcp.recv1.num + wo->tcp.recv2.num;
}
else if( w->state == CI_TCP_STATE_UDP ) {
tq = wo->udp.tx_count + oo_atomic_read(&wo->udp.tx_async_q_level);
rq = ci_udp_recv_q_pkts(&wo->udp.recv_q);
}
log("%s %d %d "OOF_IP4PORT" "OOF_IP4PORT" %s",
citp_waitable_type_str(w), rq, tq,
OOFA_IP4PORT(sock_laddr_be32(s), sock_lport_be16(s)),
OOFA_IP4PORT(sock_raddr_be32(s), sock_rport_be16(s)),
ci_tcp_state_str(w->state));
}
}
void ci_netif_filter_for_each_match(ci_netif* ni, unsigned laddr,
unsigned lport, unsigned raddr,
unsigned rport, unsigned protocol,
int intf_i, int vlan,
int (*callback)(ci_sock_cmn*, void*),
void* callback_arg, ci_uint32* hash_out)
{
ci_netif_filter_table* tbl;
unsigned hash1, hash2 = 0;
unsigned first;
tbl = ni->filter_table;
if( hash_out != NULL )
*hash_out = tcp_hash3(tbl, laddr, lport, raddr, rport, protocol);
hash1 = tcp_hash1(tbl, laddr, lport, raddr, rport, protocol);
first = hash1;
LOG_NV(log("%s: %s %s:%u->%s:%u hash=%u:%u at=%u",
__FUNCTION__, CI_IP_PROTOCOL_STR(protocol),
ip_addr_str(laddr), (unsigned) CI_BSWAP_BE16(lport),
ip_addr_str(raddr), (unsigned) CI_BSWAP_BE16(rport),
first, tcp_hash2(tbl, laddr, lport, raddr, rport, protocol),
hash1));
while( 1 ) {
int id = tbl->table[hash1].id;
if(CI_LIKELY( id >= 0 )) {
ci_sock_cmn* s = ID_TO_SOCK(ni, id);
if( ((laddr - tbl->table[hash1].laddr) |
(lport - sock_lport_be16(s) ) |
(raddr - sock_raddr_be32(s) ) |
(rport - sock_rport_be16(s) ) |
(protocol - sock_protocol(s) )) == 0 )
if(CI_LIKELY( (s->rx_bind2dev_ifindex == CI_IFID_BAD ||
ci_sock_intf_check(ni, s, intf_i, vlan)) ))
if( callback(s, callback_arg) != 0 )
return;
}
else if( id == EMPTY )
break;
/* We defer calculating hash2 until it's needed, just to make the fast
** case that little bit faster. */
if( hash1 == first )
hash2 = tcp_hash2(tbl, laddr, lport, raddr, rport, protocol);
hash1 = (hash1 + hash2) & tbl->table_size_mask;
if( hash1 == first ) {
LOG_NV(ci_log(FN_FMT "ITERATE FULL %s:%u->%s:%u hash=%u:%u",
FN_PRI_ARGS(ni), ip_addr_str(laddr), lport,
ip_addr_str(raddr), rport, hash1, hash2));
break;
}
}
}
int ci_netif_filter_lookup(ci_netif* netif, unsigned laddr, unsigned lport,
unsigned raddr, unsigned rport, unsigned protocol)
{
unsigned hash1, hash2 = 0;
ci_netif_filter_table* tbl;
unsigned first;
ci_assert(netif);
ci_assert(ci_netif_is_locked(netif));
ci_assert(netif->filter_table);
tbl = netif->filter_table;
hash1 = tcp_hash1(tbl, laddr, lport, raddr, rport, protocol);
first = hash1;
LOG_NV(log("tbl_lookup: %s %s:%u->%s:%u hash=%u:%u at=%u",
CI_IP_PROTOCOL_STR(protocol),
ip_addr_str(laddr), (unsigned) CI_BSWAP_BE16(lport),
ip_addr_str(raddr), (unsigned) CI_BSWAP_BE16(rport),
first, tcp_hash2(tbl, laddr, lport, raddr, rport, protocol),
hash1));
while( 1 ) {
int id = tbl->table[hash1].id;
if( CI_LIKELY(id >= 0) ) {
ci_sock_cmn* s = ID_TO_SOCK(netif, id);
if( ((laddr - tbl->table[hash1].laddr) |
(lport - sock_lport_be16(s) ) |
(raddr - sock_raddr_be32(s) ) |
(rport - sock_rport_be16(s) ) |
(protocol - sock_protocol(s) )) == 0 )
return hash1;
}
if( id == EMPTY ) break;
/* We defer calculating hash2 until it's needed, just to make the fast
* case that little bit faster. */
if( hash1 == first )
hash2 = tcp_hash2(tbl, laddr, lport, raddr, rport, protocol);
hash1 = (hash1 + hash2) & tbl->table_size_mask;
if( hash1 == first ) {
LOG_E(ci_log(FN_FMT "ERROR: LOOP %s:%u->%s:%u hash=%u:%u",
FN_PRI_ARGS(netif), ip_addr_str(laddr), lport,
ip_addr_str(raddr), rport, hash1, hash2));
return -ELOOP;
}
}
return -ENOENT;
}
static void thc_dump_sockets(ci_netif* netif, oo_dump_log_fn_t log,
void* log_arg)
{
unsigned id;
for( id = 0; id < netif->state->n_ep_bufs; ++id ) {
citp_waitable_obj* wo = ID_TO_WAITABLE_OBJ(netif, id);
if( wo->waitable.state != CI_TCP_STATE_FREE ) {
citp_waitable* w = &wo->waitable;
ci_sock_cmn* s = CI_CONTAINER(ci_sock_cmn, b, w);
log(log_arg, " %s lcl="OOF_IP4PORT" rmt="OOF_IP4PORT,
citp_waitable_type_str(w),
OOFA_IP4PORT(sock_laddr_be32(s), sock_lport_be16(s)),
OOFA_IP4PORT(sock_raddr_be32(s), sock_rport_be16(s)));
}
}
}
static void citp_waitable_dump2(ci_netif* ni, citp_waitable* w, const char* pf,
oo_dump_log_fn_t logger, void* log_arg)
{
unsigned tmp;
if( CI_TCP_STATE_IS_SOCKET(w->state) ) {
ci_sock_cmn* s = CI_CONTAINER(ci_sock_cmn, b, w);
logger(log_arg, "%s%s "NT_FMT"lcl="OOF_IP4PORT" rmt="OOF_IP4PORT" %s",
pf, citp_waitable_type_str(w), NI_ID(ni), W_FMT(w),
OOFA_IP4PORT(sock_laddr_be32(s), sock_lport_be16(s)),
OOFA_IP4PORT(sock_raddr_be32(s), sock_rport_be16(s)),
ci_tcp_state_str(w->state));
}
else
logger(log_arg, "%s%s "NT_FMT, pf,
citp_waitable_type_str(w), NI_ID(ni), W_FMT(w));
if( w->state == CI_TCP_STATE_FREE || w->state == CI_TCP_STATE_AUXBUF )
return;
tmp = w->lock.wl_val;
logger(log_arg, "%s lock: %x %s%s", pf, tmp,
(tmp & OO_WAITABLE_LK_LOCKED) ? "LOCKED" : "",
(tmp & OO_WAITABLE_LK_NEED_WAKE) ? " CONTENDED": "");
logger(log_arg, "%s rx_wake=%08x%s tx_wake=%08x%s flags: "CI_SB_FLAGS_FMT,
pf,
w->sleep_seq.rw.rx,
ci_bit_test(&w->wake_request, CI_SB_FLAG_WAKE_RX_B) ? "(RQ)":" ",
w->sleep_seq.rw.tx,
ci_bit_test(&w->wake_request, CI_SB_FLAG_WAKE_TX_B) ? "(RQ)":" ",
CI_SB_FLAGS_PRI_ARG(w));
if( w->spin_cycles == -1 )
logger(log_arg, "%s ul_poll: -1 spin cycles -1 usecs", pf);
else
logger(log_arg, "%s ul_poll: %llu spin cycles %u usec", pf,
w->spin_cycles, oo_cycles64_to_usec(ni, w->spin_cycles));
}
void ci_netif_filter_dump(ci_netif* ni)
{
int id;
unsigned i;
ci_netif_filter_table* tbl;
ci_assert(ni);
tbl = ni->filter_table;
log("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++");
#if CI_CFG_STATS_NETIF
log(FN_FMT "size=%d n_entries=%i n_slots=%i max=%i mean=%i", FN_PRI_ARGS(ni),
tbl->table_size_mask + 1, ni->state->stats.table_n_entries,
ni->state->stats.table_n_slots, ni->state->stats.table_max_hops,
ni->state->stats.table_mean_hops);
#endif
for( i = 0; i <= tbl->table_size_mask; ++i ) {
id = tbl->table[i].id;
if( CI_LIKELY(id >= 0) ) {
ci_sock_cmn* s = ID_TO_SOCK(ni, id);
unsigned laddr = tbl->table[i].laddr;
int lport = sock_lport_be16(s);
unsigned raddr = sock_raddr_be32(s);
int rport = sock_rport_be16(s);
int protocol = sock_protocol(s);
unsigned hash1 = tcp_hash1(tbl, laddr, lport, raddr, rport, protocol);
unsigned hash2 = tcp_hash2(tbl, laddr, lport, raddr, rport, protocol);
log("%010d id=%-10d rt_ct=%d %s "CI_IP_PRINTF_FORMAT":%d "
CI_IP_PRINTF_FORMAT":%d %010d:%010d",
i, id, tbl->table[i].route_count, CI_IP_PROTOCOL_STR(protocol),
CI_IP_PRINTF_ARGS(&laddr), CI_BSWAP_BE16(lport),
CI_IP_PRINTF_ARGS(&raddr), CI_BSWAP_BE16(rport), hash1, hash2);
}
}
log("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++");
}
static int
efab_tcp_helper_get_info(ci_private_t *unused, void *arg)
{
ci_netif_info_t *info = arg;
int index, rc=0;
tcp_helper_resource_t* thr = NULL;
ci_netif* ni = NULL;
int flags = EFAB_THR_TABLE_LOOKUP_CHECK_USER | EFAB_THR_TABLE_LOOKUP_NO_WARN;
#if CI_CFG_EFAB_EPLOCK_RECORD_CONTENTIONS
int j;
eplock_resource_t* eplock_rs;
#endif
info->ni_exists = 0;
info->ni_no_perms_exists = 0;
if( info->ni_orphan ) {
flags |= EFAB_THR_TABLE_LOOKUP_NO_UL;
info->ni_orphan = 0;
}
rc = efab_thr_table_lookup(NULL, info->ni_index, flags, &thr);
if( rc == 0 ) {
info->ni_exists = 1;
info->ni_orphan = (thr->k_ref_count & TCP_HELPER_K_RC_NO_USERLAND);
ni = &thr->netif;
info->mmap_bytes = thr->mem_mmap_bytes;
info->k_ref_count = thr->k_ref_count;
info->rs_ref_count = oo_atomic_read(&thr->ref_count);
memcpy(info->ni_name, ni->state->name, sizeof(ni->state->name));
} else if( rc == -EACCES ) {
info->ni_no_perms_id = info->ni_index;
if( efab_thr_get_inaccessible_stack_info(info->ni_index,
&info->ni_no_perms_uid,
&info->ni_no_perms_euid,
&info->ni_no_perms_share_with,
info->ni_no_perms_name) == 0 )
info->ni_no_perms_exists = 1;
}
/* sub-ops that do not need the netif to exist */
if( info->ni_subop == CI_DBG_NETIF_INFO_GET_NEXT_NETIF ) {
tcp_helper_resource_t* next_thr;
info->u.ni_next_ni.index = -1;
for( index = info->ni_index + 1;
index < 10000 /* FIXME: magic! */;
++index ) {
rc = efab_thr_table_lookup(NULL, index, flags, &next_thr);
if( rc == 0 ) {
if( next_thr->k_ref_count & TCP_HELPER_K_RC_NO_USERLAND )
efab_tcp_helper_k_ref_count_dec(next_thr, 1);
else
efab_thr_release(next_thr);
info->u.ni_next_ni.index = index;
break;
}
if( rc == -EACCES ) {
info->u.ni_next_ni.index = index;
break;
}
}
rc = 0;
}
else if( info->ni_subop == CI_DBG_NETIF_INFO_NOOP ) {
rc = 0;
}
if (!info->ni_exists)
return 0;
/* sub-ops that need the netif to exist */
switch (info->ni_subop)
{
case CI_DBG_NETIF_INFO_GET_ENDPOINT_STATE:
index = info->u.ni_endpoint.index;
info->u.ni_endpoint.max = thr->netif.ep_tbl_n;
if ((index < 0) || (index >= (int)thr->netif.ep_tbl_n)) {
info->u.ni_endpoint.state = CI_TCP_STATE_FREE;
}
else {
citp_waitable_obj* wo = ID_TO_WAITABLE_OBJ(ni, index);
info->u.ni_endpoint.state = wo->waitable.state;
if( wo->waitable.state == CI_TCP_STATE_UDP ) {
ci_udp_state* us = &wo->udp;
info->u.ni_endpoint.udpstate = us->udpflags;
info->u.ni_endpoint.rx_pkt_ul = us->recv_q.pkts_delivered;
info->u.ni_endpoint.rx_pkt_kn = us->stats.n_rx_os;
}
else if( wo->waitable.state & CI_TCP_STATE_TCP_CONN ) {
ci_tcp_state* ts = &wo->tcp;
info->u.ni_endpoint.tx_pkts_max = ts->so_sndbuf_pkts;
info->u.ni_endpoint.tx_pkts_num = ts->send.num;
}
if( wo->waitable.state & CI_TCP_STATE_SOCKET ) {
ci_sock_cmn* s = &wo->sock;
info->u.ni_endpoint.protocol = (int) sock_protocol(s);
info->u.ni_endpoint.laddr = sock_laddr_be32(s);
info->u.ni_endpoint.lport = (int) sock_lport_be16(s);
info->u.ni_endpoint.raddr = sock_raddr_be32(s);
info->u.ni_endpoint.rport = (int) sock_rport_be16(s);
}
}
break;
case CI_DBG_NETIF_INFO_GET_NEXT_NETIF:
/* If the current netif is found, we need to succeed */
break;
case CI_DBG_NETIF_INFO_NOOP:
/* Always succeeds, rc already set */
break;
default:
rc = -EINVAL;
break;
}
if( thr ) {
/* Lookup needs a matching efab_thr_release() in case of ordinary
* stack but just a ref_count_dec in case of orphan
*/
if( thr->k_ref_count & TCP_HELPER_K_RC_NO_USERLAND )
efab_tcp_helper_k_ref_count_dec(thr, 1);
else
efab_thr_release(thr);
}
return rc;
}
/* In this bind handler we just check that the address to which
* are binding is either "any" or one of ours.
* In the Linux kernel version [fd] is unused.
*/
int ci_tcp_bind(citp_socket* ep, const struct sockaddr* my_addr,
socklen_t addrlen, ci_fd_t fd )
{
struct sockaddr_in* my_addr_in;
ci_uint16 new_port;
ci_uint32 addr_be32;
ci_sock_cmn* s = ep->s;
ci_tcp_state* c = &SOCK_TO_WAITABLE_OBJ(s)->tcp;
int rc;
CHECK_TEP(ep);
my_addr_in = (struct sockaddr_in*) my_addr;
/* Check if state of the socket is OK for bind operation. */
/* \todo Earlier (TS_TCP( epi->tcpep.state )->tcp_source_be16) is used.
* What is better? */
if (my_addr == NULL)
RET_WITH_ERRNO( EINVAL );
if (s->b.state != CI_TCP_CLOSED)
RET_WITH_ERRNO( EINVAL );
if (c->tcpflags & CI_TCPT_FLAG_WAS_ESTAB)
RET_WITH_ERRNO( EINVAL );
if( my_addr->sa_family != s->domain )
RET_WITH_ERRNO( s->domain == PF_INET ? EAFNOSUPPORT : EINVAL );
/* Bug 4884: Windows regularly uses addrlen > sizeof(struct sockaddr_in)
* Linux is also relaxed about overlength data areas. */
if (s->domain == PF_INET && addrlen < sizeof(struct sockaddr_in))
RET_WITH_ERRNO( EINVAL );
#if CI_CFG_FAKE_IPV6
if (s->domain == PF_INET6 && addrlen < SIN6_LEN_RFC2133)
RET_WITH_ERRNO( EINVAL );
if( s->domain == PF_INET6 && !ci_tcp_ipv6_is_ipv4(my_addr) )
return CI_SOCKET_HANDOVER;
#endif
addr_be32 = ci_get_ip4_addr(s->domain, my_addr);
/* Using the port number provided, see if we can do this bind */
new_port = my_addr_in->sin_port;
if( CITP_OPTS.tcp_reuseports != 0 && new_port != 0 ) {
struct ci_port_list *force_reuseport;
CI_DLLIST_FOR_EACH2(struct ci_port_list, force_reuseport, link,
(ci_dllist*)(ci_uintptr_t)CITP_OPTS.tcp_reuseports) {
if( force_reuseport->port == new_port ) {
int one = 1;
ci_fd_t os_sock = ci_get_os_sock_fd(ep, fd);
ci_assert(CI_IS_VALID_SOCKET(os_sock));
rc = ci_sys_setsockopt(os_sock, SOL_SOCKET, SO_REUSEPORT, &one,
sizeof(one));
ci_rel_os_sock_fd(os_sock);
if( rc != 0 && errno == ENOPROTOOPT )
ep->s->s_flags |= CI_SOCK_FLAG_REUSEPORT_LEGACY;
ep->s->s_flags |= CI_SOCK_FLAG_REUSEPORT;
LOG_TC(log("%s "SF_FMT", applied legacy SO_REUSEPORT flag for port %u",
__FUNCTION__, SF_PRI_ARGS(ep, fd), new_port));
}
}
}
if( !(ep->s->s_flags & CI_SOCK_FLAG_REUSEPORT_LEGACY) )
CI_LOGLEVEL_TRY_RET(LOG_TV,
__ci_bind(ep->netif, ep->s, addr_be32, &new_port));
ep->s->s_flags |= CI_SOCK_FLAG_BOUND;
sock_lport_be16(s) = new_port;
sock_laddr_be32(s) = addr_be32;
if( CI_IP_IS_MULTICAST(addr_be32) )
s->cp.ip_laddr_be32 = 0;
else
s->cp.ip_laddr_be32 = addr_be32;
s->cp.lport_be16 = new_port;
sock_rport_be16(s) = sock_raddr_be32(s) = 0;
LOG_TC(log(LPF "bind to %s:%u n_p:%u lp:%u", ip_addr_str(addr_be32),
(unsigned) CI_BSWAP_BE16(my_addr_in->sin_port),
CI_BSWAP_BE16(new_port), CI_BSWAP_BE16(sock_lport_be16(s))));
return 0;
}
