static inline int init_proper_exit(void)
{
struct sigaction action = { .sa_handler = sig_handler, };
sigemptyset(&action.sa_mask);
atexit(free_resources);
if (sigaction(SIGINT, &action, NULL)
|| sigaction(SIGTERM, &action, NULL))
ERR_RET(-1, "could not install signal handlers");
return 0;
}
int load_cfg_http_svr(json_t *root, const char *key, http_svr_cfg *cfg)
{
json_t *node = json_object_get(root, key);
if (!node || !json_is_object(node))
return -__LINE__;
json_t *bind = json_object_get(node, "bind");
if (!bind)
return -__LINE__;
if (json_is_string(bind)) {
cfg->bind_count = 1;
cfg->bind_arr = malloc(sizeof(nw_svr_bind));
if (nw_sock_cfg_parse(json_string_value(bind), &cfg->bind_arr[0].addr, &cfg->bind_arr[0].sock_type) < 0)
return -__LINE__;
} else if (json_is_array(bind)) {
cfg->bind_count = json_array_size(bind);
if (cfg->bind_count == 0)
return -__LINE__;
cfg->bind_arr = malloc(sizeof(nw_svr_bind) * cfg->bind_count);
for (uint32_t i = 0; i < cfg->bind_count; ++i) {
json_t *row = json_array_get(bind, i);
if (!json_is_string(row))
return -__LINE__;
if (nw_sock_cfg_parse(json_string_value(row), &cfg->bind_arr[i].addr, &cfg->bind_arr[i].sock_type) < 0)
return -__LINE__;
}
} else {
return -__LINE__;
}
ERR_RET(read_cfg_uint32(node, "max_pkg_size", &cfg->max_pkg_size, true, 0));
ERR_RET(read_cfg_uint32(node, "buf_limit", &cfg->buf_limit, false, 0));
ERR_RET(read_cfg_uint32(node, "read_mem", &cfg->read_mem, false, 0));
ERR_RET(read_cfg_uint32(node, "write_mem", &cfg->write_mem, false, 0));
ERR_RET(read_cfg_int(node, "keep_alive", &cfg->keep_alive, false, 3600));
return 0;
}
static int init(int argc, char **argv, struct animation *banner)
{
int i;
struct string_list *filenames = NULL, *filenames_tail = NULL;
int filenames_count = 0;
init_log();
i = get_options(argc, argv);
if (i < 0)
return usage(argv[0], NULL);
for ( ; i < argc; ++i) {
if (banner->interval == (unsigned int)-1)
if (!parse_interval(argv[i], &banner->interval))
continue;
filenames_tail = string_list_add(&filenames, filenames_tail,
argv[i]);
if (!filenames_tail)
return 1;
else
filenames_count++;
}
if (!filenames_count)
return usage(argv[0], "No filenames specified");
if (fb_init(&_Fb))
return 1;
if (init_proper_exit())
return 1;
if (animation_init(filenames, filenames_count, &_Fb, banner))
return 1;
string_list_destroy(filenames);
if (banner->frame_count == 1 && RunCount == 1)
banner->interval = 0; /* Single frame, exit after showing it */
else if (banner->interval == (unsigned int)-1)
banner->interval = 1000 / 24; /* 24fps */
if (!Interactive && daemonify())
ERR_RET(1, "could not create a daemon");
return 0;
}
int
tcp_stream_get_data(tcp_stream_t *t, u_int8_t *buf, size_t size, int dir)
{
int tot;
tcp_pktq_t *q;
if (unlikely(!t || !buf || size < 1))
ERR_RET(-1, "invalid argument\n");
q = &t->pktqs[dir % 2];
tot = q->len > size ? size : q->len;
if (tot == 0)
return 0;
return _tcp_pktq_get_data(q, buf, tot);
}
int load_cfg_inetv4_list(json_t *root, const char *key, inetv4_list *cfg)
{
json_t *node = json_object_get(root, key);
if (!node || !json_is_array(node))
return -__LINE__;
cfg->count = json_array_size(node);
cfg->arr = malloc(sizeof(struct sockaddr_in) * cfg->count);
for (size_t i = 0; i < cfg->count; ++i) {
json_t *row = json_array_get(node, i);
if (!json_is_string(row)) {
return -__LINE__;
}
ERR_RET(parse_inetv4_addr(json_string_value(row), &cfg->arr[i]));
}
return 0;
}
int load_cfg_log(json_t *root, const char *key, log_cfg *cfg)
{
json_t *node = json_object_get(root, key);
if (!node || !json_is_object(node))
return -__LINE__;
ERR_RET(read_cfg_str(node, "path", &cfg->path, NULL));
ERR_RET(read_cfg_str(node, "flag", &cfg->flag, NULL));
cfg->shift = 0;
char *shift;
ERR_RET(read_cfg_str(node, "shift", &shift, "size"));
strtolower(shift);
if (strcmp(shift, "size") == 0) {
cfg->shift |= DLOG_SHIFT_BY_SIZE;
} else if (strcmp(shift, "hour") == 0) {
cfg->shift |= DLOG_SHIFT_BY_HOUR;
} else if (strcmp(shift, "min") == 0) {
cfg->shift |= DLOG_SHIFT_BY_MIN;
} else {
cfg->shift |= DLOG_SHIFT_BY_DAY;
}
bool is_pid;
bool is_fork;
ERR_RET(read_cfg_bool(node, "pid", &is_pid, false, true));
ERR_RET(read_cfg_bool(node, "fork", &is_fork, false, true));
if (is_pid) {
cfg->shift |= DLOG_LOG_PID;
}
if (is_fork) {
cfg->shift |= DLOG_USE_FORK;
}
ERR_RET(read_cfg_int(node, "max", &cfg->max, false, 100 * 1000 * 1000));
ERR_RET(read_cfg_int(node, "num", &cfg->num, false, 100));
ERR_RET(read_cfg_int(node, "keep", &cfg->keep, false, 7));
return 0;
}
int load_cfg_mysql(json_t *root, const char *key, mysql_cfg *cfg)
{
json_t *node = json_object_get(root, key);
if (!node || !json_is_object(node))
return -__LINE__;
ERR_RET(read_cfg_str(node, "host", &cfg->host, NULL));
ERR_RET(read_cfg_int(node, "port", &cfg->port, false, 3306));
ERR_RET(read_cfg_str(node, "user", &cfg->user, NULL));
ERR_RET(read_cfg_str(node, "pass", &cfg->pass, NULL));
ERR_RET(read_cfg_str(node, "name", &cfg->name, NULL));
ERR_RET(read_cfg_str(node, "charset", &cfg->charset, "utf8"));
return 0;
}
WTDB_ERR
DbSessionManager::acquire_session(DbSession **session, bool pooled, unsigned long maxWait)
{
JET_ERR rc;
if (!pooled)
{
ERR_RET( _engine->create_session(session) )
}
else
{
rc = WaitForSingleObject(_sema, maxWait);
if (rc == WAIT_TIMEOUT)
{
throw wyki_resource_timeout_error(L"Timed out waiting to acquire rights to create new database session");
}
ERR_RET( _engine->create_session(session) )
}
return JET_errSuccess;
}
objpool_t *
objpool_alloc(size_t objsize, int incsize, int locked)
{
objpool_t *op = NULL;
pthread_mutexattr_t attr;
if (objsize < 1)
return NULL;
if (incsize < 1)
incsize = OBJPOOL_INC_SIZE;
op = malloc(sizeof(objpool_t));
if (!op)
ERR_RET(NULL, "malloc objpool failed: %s\n", ERRSTR);
memset(op, 0, sizeof(objpool_t));
op->magic = time(NULL);
op->objsize = align_num(objsize);
op->incsize = incsize;
op->nodesize = op->objsize + align_num(sizeof(_op_node_t));
/* using create time as magic number, so it's unique */
if (_op_alloc_cache(op)) {
free(op);
return NULL;
}
/* initiate pthread lock for thread-safe, mutex is error-check type */
if (locked) {
op->need_lock = 1;
pthread_mutexattr_init(&attr);
pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK_NP);
pthread_mutex_init(&op->lock, &attr);
pthread_mutexattr_destroy(&attr);
}
return op;
}
int load_cfg_rpc_clt(json_t *root, const char *key, rpc_clt_cfg *cfg)
{
json_t *node = json_object_get(root, key);
if (!node || !json_is_object(node))
return -__LINE__;
ERR_RET(read_cfg_str(node, "name", &cfg->name, NULL));
json_t *addr = json_object_get(node, "addr");
if (json_is_string(addr)) {
cfg->addr_count = 1;
cfg->addr_arr = malloc(sizeof(nw_addr_t));
if (nw_sock_cfg_parse(json_string_value(addr), &cfg->addr_arr[0], &cfg->sock_type) < 0)
return -__LINE__;
} else if (json_is_array(addr)) {
cfg->addr_count = json_array_size(addr);
if (cfg->addr_count == 0)
return -__LINE__;
cfg->addr_arr = malloc(sizeof(nw_addr_t) * cfg->addr_count);
for (uint32_t i = 0; i < cfg->addr_count; ++i) {
json_t *row = json_array_get(addr, i);
if (!json_is_string(row))
return -__LINE__;
if (nw_sock_cfg_parse(json_string_value(row), &cfg->addr_arr[i], &cfg->sock_type) < 0)
return -__LINE__;
}
} else {
return -__LINE__;
}
ERR_RET(read_cfg_uint32(node, "max_pkg_size", &cfg->max_pkg_size, true, 0));
ERR_RET(read_cfg_uint32(node, "buf_limit", &cfg->buf_limit, false, 0));
ERR_RET(read_cfg_uint32(node, "read_mem", &cfg->read_mem, false, 0));
ERR_RET(read_cfg_uint32(node, "write_mem", &cfg->read_mem, false, 0));
ERR_RET(read_cfg_real(node, "reconnect_timeout", &cfg->reconnect_timeout, false, 0));
ERR_RET(read_cfg_real(node, "heartbeat_timeout", &cfg->heartbeat_timeout, false, 0));
return 0;
}
static void handle_su_serv_recv(fe_t * fe)
{
int ret;
char ipbuff[INET6_ADDRSTRLEN];
int port;
SAUN saddr;
socklen_t socklen;
su_serv_t *psvr = container_of(fe, su_serv_t, fe);
struct iovec iovrecv[2] = {{0}}; /* assumed init to 0 */
struct msghdr msgrecv = {0}; /* assumed init to 0 */
frames_t *frame;
recvagain:
socklen = sizeof(SA6);
frame = calloc(1, sizeof(frames_t) + REALDATAMAX);
if (frame == 0) {
errno = ENOBUFS; // ENOMEM
err_ret("serv %x ENOBUFS", psvr);
/* reject datagram */
ret = recvfrom(fe->fd, rejectbuff, sizeof(rejectbuff), 0, (SA*)&saddr, &socklen);
if (ret < 0 && errno == EAGAIN) {
return;
}
#ifdef SU_DEBUG_PEER_RECV
switch (saddr.sfamily) {
case PF_INET:
case PF_INET6:
#ifdef SU_DEBUG_IP6FULL
su_get_ip_port_f(&saddr, ipbuff, sizeof(ipbuff), &port);
#else
su_get_ip_port(&saddr, ipbuff, sizeof(ipbuff), &port);
#endif
break;
default:
log_msg("serv %x reject unknown protocol raw bytes %d", psvr, ret);
goto recvagain;
};
ERR_RET("serv %x %d recv %s:%d bytes %d, but reject datas", psvr,
fe->fd, ipbuff, port, ret);
#endif
return;
}
frame->srclen = psvr->servlen;
msgrecv.msg_name = & frame->srcaddr;
msgrecv.msg_namelen = frame->srclen;
msgrecv.msg_iov = iovrecv;
msgrecv.msg_iovlen = 2;
iovrecv[0].iov_base = & frame->recvhdr;
iovrecv[0].iov_len = sizeof(suhdr_t);
iovrecv[1].iov_base = frame->data;
iovrecv[1].iov_len = REALDATAMAX;
if ((ret = recvmsg(fe->fd, &msgrecv, 0)) < 0) {
if (ret < 0 && errno == EAGAIN) {
free(frame);
return;
}
ERR_RET("recvmsg error");
}
switch (frame->srcaddr.sfamily) {
case PF_INET:
case PF_INET6:
#ifdef SU_DEBUG_IP6FULL
su_get_ip_port_f(&frame->srcaddr, ipbuff, sizeof(ipbuff), &port);
#else
su_get_ip_port(&frame->srcaddr, ipbuff, sizeof(ipbuff), &port);
#endif
break;
default:
log_msg("serv %x reject unknown protocol raw bytes %d", psvr, ret);
free(frame);
goto recvagain;
};
if (ret < sizeof(suhdr_t)) {
#ifdef SU_DEBUG_PEER_RECV
errno = EBADMSG;
err_ret("serv %x recv %s:%d raw bytes %d less-then head %d bytes",
psvr, ipbuff, port, ret, sizeof(suhdr_t));
#endif
free(frame);
goto recvagain;
}
#ifdef SU_DEBUG_PEER_RECV
log_msg("serv %x recv %s:%d raw bytes %d", psvr,
ipbuff, port, ret);
#endif
suhdr_t *r = &frame->recvhdr;
uint8_t act = r->act;
uint8_t type = r->type;
frame->len = ret - sizeof(suhdr_t);
pthread_mutex_lock(&psvr->lock);
if (act == SU_SYN && frame->len > 0) {
if (!psvr->run) {
pthread_mutex_unlock(&psvr->lock);
free(frame);
goto recvagain;
}
#if defined SU_DEBUG_PEER_RECV || defined SU_DEBUG_LIST
log_msg("serv %x append syn "ColorRed"%p"ColorEnd" seq %d datagram len %d",
psvr, frame, r->seq, frame->len);
#endif
list_append(&psvr->synrecvls, &frame->node);
pthread_cond_broadcast(&psvr->syncond);
} else if (act == SU_ACK && type == SU_RELIABLE) {
if (psvr->ackwaitnum <= 0) {
pthread_mutex_unlock(&psvr->lock);
free(frame);
goto recvagain;
}
#if defined SU_DEBUG_PEER_RECV || defined SU_DEBUG_LIST
log_msg("serv %x append ack "ColorRed"%p"ColorEnd" seq %d datagram len %d",
psvr, frame, r->seq, frame->len);
#endif
list_append(&psvr->ackrecvls, &frame->node);
pthread_cond_broadcast(&psvr->ackcond);
} else {
pthread_mutex_unlock(&psvr->lock);
#ifdef SU_DEBUG_PEER_RECV
errno = EPROTO;
err_ret("serv %x recv %s:%d raw bytes %d", psvr,
ipbuff, port, ret);
#endif
free(frame);
return;
}
pthread_mutex_unlock(&psvr->lock);
goto recvagain;
}
SXSocketRef SXCreateServerSocket(unsigned short port,
int domain,
int type,
int protocol,
SXError * err_ret)
{
SXSocketRef sockPtr = (SXSocketRef)sx_calloc(1, sizeof(sx_socket_t));
if (sockPtr == NULL)
ERR_RET(SX_ERROR_MEM_ALLOC);
sockPtr->domain = domain;
sockPtr->type = type;
sockPtr->protocol = protocol;
sockPtr->ref_count = 1;
sockPtr->port = port;
socklen_t addrlen;
memset(&sockPtr->addr, 0, sizeof(struct sockaddr_storage));
if ((sockPtr->sockfd = socket(domain, type, protocol)) == -1) {
perror("socket");
ERR_RET(SX_ERROR_SYS_SOCKET);
}
switch (domain) {
case AF_INET:
sockaddr_in(sockPtr->addr).sin_addr.s_addr = INADDR_ANY;
sockaddr_in(sockPtr->addr).sin_port = htons(port);
sockaddr_in(sockPtr->addr).sin_len = sizeof(struct sockaddr_in);
addrlen = sizeof(struct sockaddr_in);
break;
case AF_INET6:
sockaddr_in6(sockPtr->addr).sin6_addr = in6addr_any;
sockaddr_in6(sockPtr->addr).sin6_port = htons(port);
sockaddr_in6(sockPtr->addr).sin6_len = sizeof(struct sockaddr_in6);
addrlen = sizeof(struct sockaddr_in6);
break;
default:
ERR_RET(SX_ERROR_INVALID_IPADDR);
return NULL;
}
sockPtr->addr.ss_family = domain;
int yes=1;
if (setsockopt(sockPtr->sockfd,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(int)) == -1) {
perror("setsockopt");
ERR_RET(SX_ERROR_SYS_SETSOCKOPT);
}
if (bind(sockPtr->sockfd, (struct sockaddr *)&sockPtr->addr, addrlen) == -1) {
perror("bind");
ERR_RET(SX_ERROR_SYS_BIND);
}
ERR_RET(SX_SUCCESS);
return sockPtr;
}
/**
* proposer_ack_redirect - Resolve an acceptor's claim that we are not the
* true proposer.
*
* If we send a prepare to an acceptor who does not believe us to be the
* true proposer, the acceptor will respond with a redirect. Since the
* correctness of Paxos guarantees that the acceptor list has a consistent
* total ordering, receiving a redirect means that there is someone more
* fitting to be proposer who we have lost contact with.
*
* Note that this does not necessarily mean that the identified proposer is
* still live; it is possible that we noticed a proposer failure and then
* prepared before the acceptor who sent the redirect detected the failure.
* To avoid this as much as possible, we wait for a majority of redirects
* before accepting defeat and attempting reconnection to our superior. If
* we "win" with a majority completing the prepare, then we drop the former
* proposer regardless of whether he has some connections still open.
*/
int
proposer_ack_redirect(struct paxos_header *hdr, msgpack_object *o)
{
int r;
struct paxos_header orig_hdr;
struct paxos_acceptor *acc;
struct paxos_continuation *k;
// We dispatched as the proposer, so we do not need to check again whether
// we think ourselves to be the proposer. Instead, just sanity check that
// the supposed true proposer has a lower ID than we do. This should
// always be the case because of the consistency of proposer ranks.
assert(hdr->ph_inum < pax->self_id);
// If we are not still preparing, either we succeeded or our prepare was
// rejected. In the former case, we should ignore the redirect because
// we have affirmed our proposership with a majority vote. In the latter
// case, if we connected to the true proposer, we would have dispatched
// as an acceptor; and if we did not successfully connect, we would have
// sent out another prepare. Hence, if we are not preparing, our prepare
// succeeded and hence we should ignore the redirect.
if (pax->prep == NULL) {
return 0;
}
// Ensure that the redirect is for our current prepare; otherwise ignore.
paxos_header_unpack(&orig_hdr, o);
if (ballot_compare(orig_hdr.ph_ballot, pax->prep->pp_ballot) != 0) {
return 0;
}
// Acknowledge the rejection of our prepare.
pax->prep->pp_redirects++;
// If we have been redirected by a majority, attempt reconnection. If a
// majority redirects, our prepare will never succeed, but we defer freeing
// it until reconnection occurs. This provides us with the guarantee (used
// above) that an acceptor who identifies as the proposer and whose prepare
// is non-NULL has either successfully prepared or has not yet begun its
// prepare cycle.
if (DEATH_ADJUSTED(pax->prep->pp_redirects) >= majority()) {
// Connect to the higher-ranked acceptor indicated in the most recent
// redirect message we received (i.e., this one). It's possible that an
// even higher-ranked acceptor exists, but we'll find that out when we
// try to send a request.
acc = acceptor_find(&pax->alist, hdr->ph_inum);
assert(acc->pa_peer == NULL);
// Defer computation until the client performs connection. If it succeeds,
// give up the prepare; otherwise, reprepare.
k = continuation_new(continue_ack_redirect, acc->pa_paxid);
ERR_RET(r, state.connect(acc->pa_desc, acc->pa_size, &k->pk_cb));
return 0;
}
// If we have heard back from everyone but the acks and redirects are tied,
// just prepare again.
if (pax->prep->pp_acks < majority() &&
DEATH_ADJUSTED(pax->prep->pp_redirects) < majority() &&
pax->prep->pp_acks + pax->prep->pp_redirects == pax->live_count) {
g_free(pax->prep);
pax->prep = NULL;
return proposer_prepare(NULL);
}
return 0;
}
/**
* paxos_commit - Commit a value for an instance of the Paxos protocol.
*
* We totally order calls to paxos_learn by instance number in order to make
* the join and greet protocols behave properly. This also gives our chat
* clients an easy mechanism for totally ordering their logs without extra
* work on their part.
*
* It is possible that failed DEC_PART decrees (i.e., decrees in which the
* proposer attempts to disconnect an acceptor who a majority of acceptors
* believe is still alive) could delay the learning of committed chat
* messages. To avoid this, once a proposer receives enough rejections
* of the decree, the part decree is replaced with a null decree. The
* proposer can then issue the part again with a higher instance number
* if desired.
*/
int
paxos_commit(struct paxos_instance *inst)
{
int r;
struct paxos_request *req = NULL;
struct paxos_instance *it;
// Mark the commit.
inst->pi_committed = true;
// Pull the request from the request cache if applicable.
if (request_needs_cached(inst->pi_val.pv_dkind)) {
req = request_find(&pax->rcache, inst->pi_val.pv_reqid);
// If we can't find a request and need one, send out a retrieve to the
// request originator and defer the commit.
if (req == NULL) {
return paxos_retrieve(inst);
}
}
// Mark the cache.
inst->pi_cached = true;
// We should already have committed and learned everything before the hole.
assert(inst->pi_hdr.ph_inum >= pax->ihole);
// Since we want our learns to be totally ordered, if we didn't just fill
// the hole, we cannot learn.
if (inst->pi_hdr.ph_inum != pax->ihole) {
// If we're the proposer, we have to just wait it out.
if (is_proposer()) {
return 0;
}
// If the hole has committed but is just waiting on a retrieve, we'll learn
// when we receive the resend.
if (pax->istart->pi_hdr.ph_inum == pax->ihole && pax->istart->pi_committed) {
assert(!pax->istart->pi_cached);
return 0;
}
// The hole is either missing or uncommitted and we are not the proposer,
// so issue a retry.
return acceptor_retry(pax->ihole);
}
// Set pax->istart to point to the instance numbered pax->ihole.
if (pax->istart->pi_hdr.ph_inum != pax->ihole) {
pax->istart = LIST_NEXT(pax->istart, pi_le);
}
assert(pax->istart->pi_hdr.ph_inum == pax->ihole);
// Now learn as many contiguous commits as we can. This function is the
// only path by which we learn commits, and we always learn in contiguous
// blocks. Therefore, it is an invariant of our system that all the
// instances numbered lower than pax->ihole are learned and committed, and
// none of the instances geq to pax->ihole are learned (although some may
// be committed).
//
// We iterate over the instance list, detecting and breaking if we find a
// hole and learning whenever we don't.
for (it = pax->istart; ; it = LIST_NEXT(it, pi_le), ++pax->ihole) {
// If we reached the end of the list, set pax->istart to the last existing
// instance.
if (it == (void *)&pax->ilist) {
pax->istart = LIST_LAST(&pax->ilist);
break;
}
// If we skipped over an instance number because we were missing an
// instance, set pax->istart to the last instance before the hole.
if (it->pi_hdr.ph_inum != pax->ihole) {
pax->istart = LIST_PREV(it, pi_le);
break;
}
// If we found an uncommitted or uncached instance, set pax->istart to it.
if (!it->pi_committed || !it->pi_cached) {
pax->istart = it;
break;
}
// By our invariant, since we are past our original hole, no instance
// should be learned.
assert(!it->pi_learned);
// Grab its associated request. This is guaranteed to exist because we
// have checked that pi_cached holds.
req = NULL;
if (request_needs_cached(it->pi_val.pv_dkind)) {
req = request_find(&pax->rcache, it->pi_val.pv_reqid);
assert(req != NULL);
}
// Learn the value.
ERR_RET(r, paxos_learn(it, req));
}
return 0;
}
int
tcp_stream_flow(tcp_stream_t *t, const netpkt_t *pkt, int dir)
{
int ret;
struct tcphdr *h;
if (unlikely(!t || !pkt))
ERR_RET(-1, "invalid argument\n");
if (unlikely(!pkt->hdr4_tcp))
ERR_RET(-1, "not decode TCP header\n");
h = pkt->hdr4_tcp;
/* check seq/ack number */
if (_tcp_check_seq(t, h, dir))
ERR_RET(-1, "invalid seq number\n");
/* RST packet */
if (h->rst) {
printf("<%d> RST: seq %u, ack %u\n",
dir, ntohl(h->seq), ntohl(h->ack));
return -1;
}
printf("<%d>stream(%p) before state: %d\n", dir, t, t->state);
/* TCP state */
switch (t->state) {
case TCP_ST_SYN:
/* server syn-ack packet */
if (dir == 1 && h->syn && h->ack)
t->state = TCP_ST_SYN_ACK;
break;
case TCP_ST_SYN_ACK:
/* client ACK, establish connection */
if (dir == 0 && h->ack)
t->state = TCP_ST_EST;
break;
case TCP_ST_EST:
/* receive Fin */
if (h->fin)
t->state = TCP_ST_FIN1;
if (h->psh) {
ret = _tcp_pktq_add(&t->pktqs[dir], pkt);
}
break;
case TCP_ST_FIN1:
if (h->ack)
t->state = TCP_ST_FIN1_ACK;
if (h->fin)
t->state = TCP_ST_FIN2;
if (h->psh) {
ret = _tcp_pktq_add(&t->pktqs[dir], pkt);
}
break;
case TCP_ST_FIN1_ACK:
if (h->fin)
t->state = TCP_ST_FIN2;
break;
case TCP_ST_FIN2:
if (h->ack) {
t->state = TCP_ST_CLOSE;
return 0;
}
break;
default:
break;
}
printf("<%d>stream(%p) after state: %d\n", dir, t, t->state);
return ret;
}
static void *thread_request_handle(void *v)
{
int ret;
struct list *synnode;
struct timespec abstime = {0};
frames_t *frame;
su_serv_t *psvr = (su_serv_t*)v;
pthread_t tid __attribute__((unused)) = pthread_self();
for (; psvr->run;) {
pthread_mutex_lock(&psvr->lock);
while ((synnode = psvr->synrecvls.next) == &psvr->synrecvls) {
maketimeout_seconds(&abstime, 1);
ret = pthread_cond_timedwait(&psvr->syncond, &psvr->lock, &abstime);
if (!psvr->run) {
pthread_mutex_unlock(&psvr->lock);
goto quit;
}
if ( ret == ETIMEDOUT ) {
pthread_mutex_lock(&psvr->cachelock);
reliable_ack_unsave(psvr);
pthread_mutex_unlock(&psvr->cachelock);
}
}
list_remove(synnode);
pthread_mutex_unlock(&psvr->lock);
/* have request datagram */
frame = container_of(synnode, frames_t, node);
rb_key_cache_t key;
memcpy(&key.destaddr, &frame->srcaddr, sizeof(SAUN));
key.destlen = frame->srclen;
key.seq = frame->recvhdr.seq;
key.sid = frame->recvhdr.sid;
struct rb_node *cachenode;
cache_t *cache;
pthread_mutex_lock(&psvr->cachelock);
reliable_ack_unsave(psvr);
if (frame->recvhdr.type == SU_RELIABLE) {
if ( (cachenode = rb_search(&psvr->rbackcache, &key))) {
cache = rb_entry(cachenode, cache_t, rbn);
if (cache->frame.len == -1) {
#ifdef SU_DEBUG_RBTREE
char ipbuff[INET6_ADDRSTRLEN];
int port;
su_get_ip_port_f(&cache->frame.srcaddr, ipbuff, sizeof(ipbuff), &port);
log_msg("serv %x %x time %u key(%s:%d:%u:%u)"
ColorRed " 0ACK cache %p" ColorEnd,
psvr, tid, cache->ts, ipbuff, port,
cache->frame.recvhdr.sid, cache->frame.recvhdr.seq, cache);
#endif
pthread_mutex_unlock(&psvr->cachelock);
free(frame);
continue;
}
#ifdef SU_DEBUG_RBTREE
char ipbuff[INET6_ADDRSTRLEN];
int port;
su_get_ip_port_f(&cache->frame.srcaddr, ipbuff, sizeof(ipbuff), &port);
log_msg("serv %x %x time %u key(%s:%d:%u:%u)"
ColorRed " @ACK cache %p" ColorEnd,
psvr, tid, cache->ts, ipbuff, port,
cache->frame.recvhdr.sid, cache->frame.recvhdr.seq, cache);
#endif
struct iovec iovsend[2] = {{0}};
struct msghdr msgsend = {0}; /* assumed init to 0 */
frame->recvhdr.act = SU_ACK;
msgsend.msg_name = (void*)&cache->frame.srcaddr;
msgsend.msg_namelen = cache->frame.srclen;
msgsend.msg_iov = &iovsend[0];
msgsend.msg_iovlen = 2;
iovsend[0].iov_base = &frame->recvhdr;
iovsend[0].iov_len = sizeof(suhdr_t);
iovsend[1].iov_base = (void*)cache->frame.data; /* get the cache results */
iovsend[1].iov_len = cache->frame.len;
/* resend from cache */
if (sendmsg(psvr->fd, &msgsend, 0) != sizeof(suhdr_t) + cache->frame.len) {
char ipbuff[INET6_ADDRSTRLEN];
int port;
su_get_ip_port_f(&cache->frame.srcaddr, ipbuff, sizeof(ipbuff), &port);
ERR_RET("retransmit sendmsg %s:%d:%u:%u:%u error",
ipbuff, port,
frame->recvhdr.seq, frame->recvhdr.ts,
frame->recvhdr.sid);
}
#ifdef SU_DEBUG_PEER_RESEND
else {
char ipbuff[INET6_ADDRSTRLEN];
int port;
su_get_ip_port_f(&cache->frame.srcaddr, ipbuff, sizeof(ipbuff), &port);
log_msg("retransmit sendmsg %s:%d:%u:%u:%u",
ipbuff, port,
frame->recvhdr.seq, frame->recvhdr.ts,
frame->recvhdr.sid);
}
#endif
pthread_mutex_unlock(&psvr->cachelock);
free(frame);
continue;
} else {
if (reliable_ack___hold(psvr, frame) < 0) {
err_ret("reliable_ack___hold error");
pthread_mutex_unlock(&psvr->cachelock);
free(frame);
continue;
}
}
}
pthread_mutex_unlock(&psvr->cachelock);
request_handle(psvr, frame);
#if defined SU_DEBUG_PEER_RECV || defined SU_DEBUG_LIST
log_msg("serv %x %x delete syn "ColorRed"%p"ColorEnd" seq %d datagram len %d",
psvr, tid, frame, frame->recvhdr.seq, frame->len);
#endif
free(frame);
}
quit:
pthread_exit(0);
}
int loop (int sock, struct sockaddr_nl *addr)
{
int received_bytes = 0;
struct nlmsghdr *nlh;
char destination_address[32];
char gateway_address[32];
struct rtmsg *route_entry; /* This struct represent a route entry \
in the routing table */
struct rtattr *route_attribute; /* This struct contain route \
attributes (route type) */
int route_attribute_len = 0;
char buffer[BUFFER_SIZE];
bzero(destination_address, sizeof(destination_address));
bzero(gateway_address, sizeof(gateway_address));
bzero(buffer, sizeof(buffer));
/* Receiving netlink socket data */
while (1)
{
received_bytes = recv(sock, buffer, sizeof(buffer), 0);
if (received_bytes < 0)
ERR_RET("recv");
/* cast the received buffer */
nlh = (struct nlmsghdr *) buffer;
/* If we received all data ---> break */
if (nlh->nlmsg_type == NLMSG_DONE)
break;
/* We are just intrested in Routing information */
if (addr->nl_groups == RTMGRP_IPV4_ROUTE)
break;
}
/* Reading netlink socket data */
/* Loop through all entries */
/* For more informations on some functions :
* http://www.kernel.org/doc/man-pages/online/pages/man3/netlink.3.html
* http://www.kernel.org/doc/man-pages/online/pages/man7/rtnetlink.7.html
*/
for ( ; NLMSG_OK(nlh, received_bytes); \
nlh = NLMSG_NEXT(nlh, received_bytes))
{
/* Get the route data */
route_entry = (struct rtmsg *) NLMSG_DATA(nlh);
/* We are just intrested in main routing table */
if (route_entry->rtm_table != RT_TABLE_MAIN)
continue;
/* Get attributes of route_entry */
route_attribute = (struct rtattr *) RTM_RTA(route_entry);
/* Get the route atttibutes len */
route_attribute_len = RTM_PAYLOAD(nlh);
/* Loop through all attributes */
for ( ; RTA_OK(route_attribute, route_attribute_len); \
route_attribute = RTA_NEXT(route_attribute, route_attribute_len))
{
/* Get the destination address */
if (route_attribute->rta_type == RTA_DST)
{
inet_ntop(AF_INET, RTA_DATA(route_attribute), \
destination_address, sizeof(destination_address));
}
/* Get the gateway (Next hop) */
if (route_attribute->rta_type == RTA_GATEWAY)
{
inet_ntop(AF_INET, RTA_DATA(route_attribute), \
gateway_address, sizeof(gateway_address));
}
}
/* Now we can dump the routing attributes */
if (nlh->nlmsg_type == RTM_DELROUTE)
fprintf(stdout, "Deleting route to destination --> %s and gateway %s\n", \
destination_address, gateway_address);
if (nlh->nlmsg_type == RTM_NEWROUTE)
printf("Adding route to destination --> %s and gateway %s\n", \
destination_address, gateway_address);
}
return 0;
}
static int su_serv_send_recv_act(su_serv_t *psvr, SA *destaddr, socklen_t destlen,
const void *outbuff, int outbytes, void *inbuff, int inbytes, int retransmit)
{
int n;
struct iovec iovsend[2]= {{0}};
struct msghdr msgsend = {0}; /* assumed init to 0 */
suhdr_t *r, sendhdr = {0}; /* protocol header */
int ret, waitsec;
struct list *node = 0;
frames_t *packet = 0;
pthread_mutex_lock(&psvr->mutex);
pthread_mutex_lock(&psvr->lock);
if (retransmit == 0) {
psvr->seq++;
psvr->retransmission = 1;
} else {
if (psvr->retransmission == 0) {
pthread_mutex_unlock(&psvr->mutex);
pthread_mutex_unlock(&psvr->lock);
errno = ETIMEDOUT;
return -1;
}
psvr->retransmission --;
}
if (psvr->rttinit == 0) {
rtt_init(&psvr->rttinfo, psvr->retry); /* first time we're called */
psvr->rttinit = 1;
}
sendhdr.act = SU_SYN;
sendhdr.type = SU_RELIABLE;
sendhdr.sid = psvr->sid;
sendhdr.seq = psvr->seq;
msgsend.msg_name = (void*)destaddr;
msgsend.msg_namelen = destlen;
msgsend.msg_iov = iovsend;
msgsend.msg_iovlen = 2;
iovsend[0].iov_base = (void*)&sendhdr;
iovsend[0].iov_len = sizeof(suhdr_t);
iovsend[1].iov_base = (void*)outbuff;
iovsend[1].iov_len = outbytes;
struct timespec abstime = {0};
suhdr_t *precvhdr;
rtt_newpack(&psvr->rttinfo); /* initialize for this packet */
psvr->ackwaitnum ++;
sendagain:
sendhdr.ts = rtt_ts(&psvr->rttinfo);
if (sendmsg(psvr->fd, &msgsend, 0) < 0) {
ERR_RET("sendmsg error");
goto error_ret;
}
waitsec = rtt_start(&psvr->rttinfo); /* calc timeout value & start timer */
#ifdef SU_DEBUG_RTT
fprintf(stderr, ColorRed "send seq %4d: " ColorEnd, sendhdr.seq);
rtt_debug(&psvr->rttinfo);
#endif
/* set timed wait time-point */
maketimeout_seconds(&abstime, waitsec);
#ifdef SU_DEBUG_TIMEVERBOSE
struct timeval now;
gettimeofday(&now, 0);
log_msg( ColorBlue "pthread_cond_timedwait : %u.%u time expire" ColorEnd,
abstime.tv_sec, abstime.tv_nsec);
log_msg( ColorBlue "pthread_cond_timedwait : %d.%d now time" ColorEnd,
now.tv_sec, now.tv_usec*1000);
#endif
timedwaitagain:
ret = pthread_cond_timedwait(&psvr->ackcond, &psvr->lock, &abstime);
if (ret == 0) {
#ifdef SU_DEBUG_TIMEVERBOSE
struct timeval now;
gettimeofday(&now, 0);
log_msg(ColorBlue "pthread_cond_timedwait : %d.%d ack cond interrupt" ColorEnd,
now.tv_sec, now.tv_usec*1000);
#endif
node = psvr->ackrecvls.next;
for (; node != &psvr->ackrecvls; node = node->next) {
packet = container_of(node, frames_t, node);
r = &packet->recvhdr;
if (su_cmp_ack_SU_RELIABLE(&sendhdr, r)) {
break;
}
}
if ( node == &psvr->ackrecvls ) {
/* Be careful of the lock, locked -> timedwait -> unlock */
#ifdef SU_DEBUG_LIST
log_msg("serv %x no found seq %d ack, timed wait again", psvr, sendhdr.seq);
#endif
goto timedwaitagain;
}
/* Find response packet node */
list_remove(&packet->node);
n = packet->len;
precvhdr = &packet->recvhdr;
#if defined SU_DEBUG_PEER_RECV || defined SU_DEBUG_LIST
log_msg("serv %x finded ack " ColorRed "%p" ColorEnd " seq %d datagram len %d",
psvr, packet, r->seq, packet->len);
#endif
#ifdef SU_DEBUG_RTT
fprintf(stderr, ColorRed "recv seq %4d \n" ColorEnd, precvhdr->seq);
#endif
// SU_RELIABLE received response, copy to user buffer
memcpy(inbuff, packet->data, n > inbytes ? inbytes : n);
} else if (ret == EINTR) {
log_msg("pthread_cond_timedwait system EINTR");
goto timedwaitagain;
} else if (ret == ETIMEDOUT) {
#ifdef SU_DEBUG_TIMEVERBOSE
struct timeval now;
gettimeofday(&now, 0);
log_msg(ColorBlue "pthread_cond_timedwait : %u.%u ETIMEOUT have expired" ColorEnd,
now.tv_sec, now.tv_usec*1000);
#endif
if (rtt_timeout(&psvr->rttinfo) < 0) {
#ifdef SU_DEBUG_RTT
err_msg(ColorYel "no response from server, giving up" ColorEnd);
#endif
psvr->rttinit = 0; /* reinit in case we're called again */
errno = ETIMEDOUT;
goto error_ret;
}
#ifdef SU_DEBUG_RTT
err_msg(ColorRed " seq %4d timeout, retransmitting %d" ColorEnd,
sendhdr.seq, ++retransmit);
#endif
goto sendagain;
} else {
errno = ret;
ERR_RET(" su_serv_send_recv_act unknown error[%d]", ret);
goto error_ret;
}
/* calculate & store new RTT estimator values */
rtt_stop(&psvr->rttinfo, rtt_ts(&psvr->rttinfo) - precvhdr->ts);
if (--psvr->ackwaitnum == 0) {
su_serv_list_empty(psvr, &psvr->ackrecvls);
}
pthread_mutex_unlock(&psvr->mutex);
pthread_mutex_unlock(&psvr->lock);
#ifdef SU_DEBUG_LIST
log_msg("serv %x free node " ColorRed "%p"ColorEnd" seq %d", psvr, packet, sendhdr.seq);
#endif
free(packet);
return(n); /* return size of received datagram */
error_ret:
if (--psvr->ackwaitnum == 0) {
su_serv_list_empty(psvr, &psvr->ackrecvls);
}
pthread_mutex_unlock(&psvr->mutex);
pthread_mutex_unlock(&psvr->lock);
return(-1);
}
static void handle_su_peer_recv(fe_t * fe)
{
int ret, port;
char ipbuff[INET6_ADDRSTRLEN];
SAUN saddr;
socklen_t socklen;
su_peer_t *psar = container_of(fe, su_peer_t, fe);
struct iovec iovrecv[2] = {{0}}; /* assumed init to 0 */
struct msghdr msgrecv = {0}; /* assumed init to 0 */
frames_t *frame;
recvagain:
socklen = psar->destlen;
frame = calloc(1, sizeof(frames_t) + REALDATAMAX);
if (frame == 0) {
errno = ENOBUFS; // ENOMEM
log_msg("peer %x ENOBUFS", psar);
/* reject datagram */
ret = recvfrom(fe->fd, rejectbuff, sizeof(rejectbuff), 0, (SA*)&saddr, &socklen);
if (ret < 0 && errno == EAGAIN) {
return;
}
#ifdef SU_DEBUG_PEER_RECV
switch (saddr.sfamily) {
case PF_INET:
case PF_INET6:
#ifdef SU_DEBUG_IP6FULL
su_get_ip_port_f(&saddr, ipbuff, sizeof(ipbuff), &port);
#else
su_get_ip_port(&saddr, ipbuff, sizeof(ipbuff), &port);
#endif
break;
default:
log_msg("peer %x reject unknown protocol raw bytes %d", psar, ret);
free(frame);
goto recvagain;
};
ERR_RET("peer %x recv %s:%d bytes %d, but reject datas",
psar, ipbuff, port, ret);
#endif
return;
}
frame->srclen = psar->destlen;
msgrecv.msg_name = & frame->srcaddr;
msgrecv.msg_namelen = frame->srclen;
msgrecv.msg_iov = iovrecv;
msgrecv.msg_iovlen = 2;
iovrecv[0].iov_base = & frame->recvhdr;
iovrecv[0].iov_len = sizeof(suhdr_t);
iovrecv[1].iov_base = frame->data;
iovrecv[1].iov_len = REALDATAMAX;
if ((ret = recvmsg(fe->fd, &msgrecv, 0)) < 0) {
if (ret < 0 && errno == EAGAIN) {
free(frame);
return;
}
ERR_RET("recvmsg error");
}
switch (frame->srcaddr.sfamily) {
case PF_INET:
case PF_INET6:
#ifdef SU_DEBUG_IP6FULL
su_get_ip_port_f(&frame->srcaddr, ipbuff, sizeof(ipbuff), &port);
#else
su_get_ip_port(&frame->srcaddr, ipbuff, sizeof(ipbuff), &port);
#endif
break;
default:
log_msg("peer %x reject unknown protocol type %d raw bytes %d",
psar, frame->srcaddr.sfamily, ret);
free(frame);
goto recvagain;
};
#ifdef SU_DEBUG_PEER_RECV
log_msg("peer %x recv %s:%d raw bytes %d", psar, ipbuff, port, ret);
#endif
if (ret < sizeof(suhdr_t)) {
#ifdef SU_DEBUG_PEER_RECV
errno = EBADMSG;
err_ret("peer %x recv %s:%d raw bytes %d less than the protocol header %d", psar,
ipbuff, port, ret, sizeof(suhdr_t));
#endif
free(frame);
goto recvagain;
}
suhdr_t *r = &frame->recvhdr;
uint8_t act = r->act;
uint8_t type = r->type;
frame->len = ret - sizeof(suhdr_t);
SAUN *psrc, *pdst;
psrc = &frame->srcaddr; // foreign host
pdst = &psar->destaddr; // localhost
#ifndef promiscuous_mode
/* Filter: Check address and port
* compare datagram source and peer destination */
if ( (pdst->sfamily == PF_INET6 &&
sockaddr_in6_cmp(&psrc->addr6, &pdst->addr6 ) != 0)
|| (pdst->sfamily == PF_INET &&
sockaddr_in4_cmp(&psrc->addr4, &pdst->addr4 ) != 0) ){
#ifdef SU_DEBUG_PEER_RECV
log_msg(ColorYel"peer %x reject act[0x%02x] from %s:%d datagram len %d"ColorEnd,
act, psar, ipbuff, port, frame->len);
#endif
free(frame);
goto recvagain;
}
#endif /* #ifndef promiscuous_mode */
pthread_mutex_lock(&psar->lock);
if (act == SU_SYN && frame->len > 0) {
if (!psar->run) {
log_msg("peer %x thread handle no run");
pthread_mutex_unlock(&psar->lock);
free(frame);
goto recvagain;
}
#if defined SU_DEBUG_PEER_RECV || defined SU_DEBUG_LIST
log_msg("peer %x append syn "ColorRed"%p"ColorEnd" seq %d datagram len %d",
psar, frame, r->seq, frame->len);
#endif
list_append(&psar->synrecvls, &frame->node);
pthread_mutex_unlock(&psar->lock);
pthread_cond_broadcast(&psar->syncond);
goto recvagain;
} else if (act == SU_ACK && type == SU_RELIABLE) {
#ifdef promiscuous_mode
/* Filter: receive response from self request */
if ( (pdst->sfamily == PF_INET6 &&
sockaddr_in6_cmp(&psrc->addr6, &pdst->addr6 ) != 0)
|| (pdst->sfamily == PF_INET &&
sockaddr_in4_cmp(&psrc->addr4, &pdst->addr4 ) != 0) ){
#ifdef SU_DEBUG_PEER_RECV
log_msg(ColorYel "peer %x reject ack from %s:%d datagram len %d" ColorEnd, psar,
ipbuff, port, frame->len);
#endif
pthread_mutex_unlock(&psar->lock);
free(frame);
goto recvagain;
}
#endif /* #ifdef promiscuous_mode */
if (psar->ackwaitnum <= 0) {
pthread_mutex_unlock(&psar->lock);
free(frame);
goto recvagain;
}
#if defined SU_DEBUG_PEER_RECV || defined SU_DEBUG_LIST
log_msg("peer %x append ack "ColorRed"%p"ColorEnd" seq %d datagram len %d",
psar, frame, r->seq, frame->len);
#endif
list_append(&psar->ackrecvls, &frame->node);
pthread_mutex_unlock(&psar->lock);
pthread_cond_broadcast(&psar->ackcond);
goto recvagain;
} else {
pthread_mutex_unlock(&psar->lock);
#ifdef SU_DEBUG_PEER_RECV
errno = EPROTO;
err_ret("peer %x recv %s:%d raw bytes %d", psar, ipbuff, port, ret);
#endif
free(frame);
return;
}
pthread_mutex_unlock(&psar->lock);
goto recvagain;
}
