int
cleanup_(Module m)
{
tcp_cleanup();
freelinklist(ztcp_sessions, (FreeFunc) ztcp_free_session);
return setfeatureenables(m, &module_features, NULL);
}
/* ARGSUSED */
void
tcp_time_wait_collector(void *arg)
{
tcp_t *tcp;
int64_t now;
mblk_t *mp;
conn_t *connp;
kmutex_t *lock;
boolean_t removed;
extern void (*cl_inet_disconnect)(netstackid_t, uint8_t, sa_family_t,
uint8_t *, in_port_t, uint8_t *, in_port_t, void *);
squeue_t *sqp = (squeue_t *)arg;
tcp_squeue_priv_t *tcp_time_wait =
*((tcp_squeue_priv_t **)squeue_getprivate(sqp, SQPRIVATE_TCP));
mutex_enter(&tcp_time_wait->tcp_time_wait_lock);
tcp_time_wait->tcp_time_wait_tid = 0;
#ifdef DEBUG
tcp_time_wait->tcp_time_wait_running = B_TRUE;
#endif
if (tcp_time_wait->tcp_free_list != NULL &&
tcp_time_wait->tcp_free_list->tcp_in_free_list == B_TRUE) {
TCP_G_STAT(tcp_freelist_cleanup);
while ((tcp = tcp_time_wait->tcp_free_list) != NULL) {
tcp_time_wait->tcp_free_list = tcp->tcp_time_wait_next;
tcp->tcp_time_wait_next = NULL;
tcp_time_wait->tcp_free_list_cnt--;
ASSERT(tcp->tcp_tcps == NULL);
CONN_DEC_REF(tcp->tcp_connp);
}
ASSERT(tcp_time_wait->tcp_free_list_cnt == 0);
}
/*
* In order to reap time waits reliably, we should use a
* source of time that is not adjustable by the user -- hence
* the call to ddi_get_lbolt64().
*/
now = ddi_get_lbolt64();
while ((tcp = tcp_time_wait->tcp_time_wait_head) != NULL) {
/*
* lbolt64 should not wrap around in practice... So we can
* do a direct comparison.
*/
if (now < tcp->tcp_time_wait_expire)
break;
removed = tcp_time_wait_remove(tcp, tcp_time_wait);
ASSERT(removed);
connp = tcp->tcp_connp;
ASSERT(connp->conn_fanout != NULL);
lock = &connp->conn_fanout->connf_lock;
/*
* This is essentially a TW reclaim fast path optimization for
* performance where the timewait collector checks under the
* fanout lock (so that no one else can get access to the
* conn_t) that the refcnt is 2 i.e. one for TCP and one for
* the classifier hash list. If ref count is indeed 2, we can
* just remove the conn under the fanout lock and avoid
* cleaning up the conn under the squeue, provided that
* clustering callbacks are not enabled. If clustering is
* enabled, we need to make the clustering callback before
* setting the CONDEMNED flag and after dropping all locks and
* so we forego this optimization and fall back to the slow
* path. Also please see the comments in tcp_closei_local
* regarding the refcnt logic.
*
* Since we are holding the tcp_time_wait_lock, its better
* not to block on the fanout_lock because other connections
* can't add themselves to time_wait list. So we do a
* tryenter instead of mutex_enter.
*/
if (mutex_tryenter(lock)) {
mutex_enter(&connp->conn_lock);
if ((connp->conn_ref == 2) &&
(cl_inet_disconnect == NULL)) {
ipcl_hash_remove_locked(connp,
connp->conn_fanout);
/*
* Set the CONDEMNED flag now itself so that
* the refcnt cannot increase due to any
* walker.
*/
connp->conn_state_flags |= CONN_CONDEMNED;
mutex_exit(lock);
mutex_exit(&connp->conn_lock);
if (tcp_time_wait->tcp_free_list_cnt <
tcp_free_list_max_cnt) {
/* Add to head of tcp_free_list */
mutex_exit(
&tcp_time_wait->tcp_time_wait_lock);
tcp_cleanup(tcp);
ASSERT(connp->conn_latch == NULL);
ASSERT(connp->conn_policy == NULL);
ASSERT(tcp->tcp_tcps == NULL);
ASSERT(connp->conn_netstack == NULL);
mutex_enter(
&tcp_time_wait->tcp_time_wait_lock);
tcp->tcp_time_wait_next =
tcp_time_wait->tcp_free_list;
tcp_time_wait->tcp_free_list = tcp;
tcp_time_wait->tcp_free_list_cnt++;
continue;
} else {
/* Do not add to tcp_free_list */
mutex_exit(
&tcp_time_wait->tcp_time_wait_lock);
tcp_bind_hash_remove(tcp);
ixa_cleanup(tcp->tcp_connp->conn_ixa);
tcp_ipsec_cleanup(tcp);
CONN_DEC_REF(tcp->tcp_connp);
}
} else {
CONN_INC_REF_LOCKED(connp);
mutex_exit(lock);
mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
mutex_exit(&connp->conn_lock);
/*
* We can reuse the closemp here since conn has
* detached (otherwise we wouldn't even be in
* time_wait list). tcp_closemp_used can safely
* be changed without taking a lock as no other
* thread can concurrently access it at this
* point in the connection lifecycle.
*/
if (tcp->tcp_closemp.b_prev == NULL)
tcp->tcp_closemp_used = B_TRUE;
else
cmn_err(CE_PANIC,
"tcp_timewait_collector: "
"concurrent use of tcp_closemp: "
"connp %p tcp %p\n", (void *)connp,
(void *)tcp);
TCP_DEBUG_GETPCSTACK(tcp->tcmp_stk, 15);
mp = &tcp->tcp_closemp;
SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
tcp_timewait_close, connp, NULL,
SQ_FILL, SQTAG_TCP_TIMEWAIT);
}
} else {
mutex_enter(&connp->conn_lock);
CONN_INC_REF_LOCKED(connp);
mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
mutex_exit(&connp->conn_lock);
/*
* We can reuse the closemp here since conn has
* detached (otherwise we wouldn't even be in
* time_wait list). tcp_closemp_used can safely
* be changed without taking a lock as no other
* thread can concurrently access it at this
* point in the connection lifecycle.
*/
if (tcp->tcp_closemp.b_prev == NULL)
tcp->tcp_closemp_used = B_TRUE;
else
cmn_err(CE_PANIC, "tcp_timewait_collector: "
"concurrent use of tcp_closemp: "
"connp %p tcp %p\n", (void *)connp,
(void *)tcp);
TCP_DEBUG_GETPCSTACK(tcp->tcmp_stk, 15);
mp = &tcp->tcp_closemp;
SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
tcp_timewait_close, connp, NULL,
SQ_FILL, SQTAG_TCP_TIMEWAIT);
}
mutex_enter(&tcp_time_wait->tcp_time_wait_lock);
}
if (tcp_time_wait->tcp_free_list != NULL)
tcp_time_wait->tcp_free_list->tcp_in_free_list = B_TRUE;
/*
* If the time wait list is not empty and there is no timer running,
* restart it.
*/
if ((tcp = tcp_time_wait->tcp_time_wait_head) != NULL &&
tcp_time_wait->tcp_time_wait_tid == 0) {
hrtime_t firetime;
firetime = TICK_TO_NSEC(tcp->tcp_time_wait_expire - now);
/* This ensures that we won't wake up too often. */
firetime = MAX(TCP_TIME_WAIT_DELAY, firetime);
tcp_time_wait->tcp_time_wait_tid =
timeout_generic(CALLOUT_NORMAL, tcp_time_wait_collector,
sqp, firetime, CALLOUT_TCP_RESOLUTION,
CALLOUT_FLAG_ROUNDUP);
}
#ifdef DEBUG
tcp_time_wait->tcp_time_wait_running = B_FALSE;
#endif
mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
}
void ip_cleanup(Slirp *slirp)
{
udp_cleanup(slirp);
tcp_cleanup(slirp);
icmp_cleanup(slirp);
}
/* Main routine for the server threads */
thr_startfunc_t serve_pipe(void *data)
{
char sio_buf[BUFSIZ], sock_buf[BUFSIZ];
int fd_max, sio_fd, sock_fd;
int sio_count, sock_count;
int res, port;
fd_set rfds, wfds;
pipe_s *pipe = (pipe_s *)data;
#if defined(__UNIX__)
struct timeval tv = {pipe->timeout, 0};
struct timeval *ptv = &tv;
#elif defined(__WIN32__)
struct timeval tv = {0,10000};
struct timeval *ptv = &tv;
DWORD msecs = 0, timeout = pipe->timeout * 1000;
#endif
port = pipe->sio.info.port;
/* Only proceed if we can lock the mutex */
if (thr_mutex_trylock(pipe->mutex))
{
error("server(%d) - resource is locked", port);
}
else
{
sio_count = 0;
sock_count = 0;
sio_fd = pipe->sio.fd;
sock_fd = pipe->sock.fd;
#if defined(__UNIX__)
fd_max = sio_fd > sock_fd ? sio_fd : sock_fd;
#elif defined(__WIN32__)
fd_max = sock_fd;
msecs = GetTickCount();
#endif
fprintf(stderr, "server(%d) - thread started\n", port);
while (1)
{
FD_ZERO(&rfds);
FD_ZERO(&wfds);
#if defined(__UNIX__)
/* Always ask for read notification to check for EOF */
FD_SET(sio_fd, &rfds);
/* Only ask for write notification if we have something to write */
if (sock_count > 0)
FD_SET(sio_fd, &wfds);
/* Reset timeout values */
tv.tv_sec = pipe->timeout;
tv.tv_usec = 0;
#endif
/* Always ask for read notification to check for EOF */
FD_SET(sock_fd, &rfds);
/* Only ask for write notification if we have something to write */
if (sio_count > 0)
FD_SET(sock_fd, &wfds);
//DBG_MSG2("server(%d) waiting for events", port);
/* Wait for read/write events */
res = select(fd_max + 1, &rfds, &wfds, NULL, ptv);
if (res == -1)
{
perror2("server(%d) - select()", port);
break;
}
#if defined(__UNIX__)
/* Use the select result for timeout detection */
if (res == 0)
{
fprintf(stderr, "server(%d) - timed out\n", port);
break;
}
/* Input from serial port? */
if (FD_ISSET(sio_fd, &rfds))
#elif defined(__WIN32__)
if (1)
#endif
{
/* Only read input if buffer is empty */
if (sio_count == 0)
{
sio_count = sio_read(&pipe->sio, sio_buf, sizeof(sio_buf));
if (sio_count <= 0)
{
if (sio_count == 0)
{
#if defined(__UNIX__)
fprintf(stderr, "server(%d) - EOF from sio\n", port);
break;
#endif
}
else
{
perror2("server(%d) - read(sio)", port);
break;
}
}
else
{
DBG_MSG3("server(%d) - read %d bytes from sio", port, sio_count);
}
}
}
/* Write to socket possible? */
if (FD_ISSET(sock_fd, &wfds))
{
if (sio_count > 0)
{
if ((res = tcp_write(&pipe->sock, sio_buf, sio_count)) < 0)
{
perror2("server(%d) - write(sock)", port);
break;
}
DBG_MSG3("server(%d) - Wrote %d bytes to sock", port, res);
sio_count -= res;
}
}
/* Input from socket? */
if (FD_ISSET(sock_fd, &rfds))
{
/* Only read input if buffer is empty */
if (sock_count == 0)
{
sock_count = tcp_read(&pipe->sock, sock_buf, sizeof(sock_buf));
if (sock_count <= 0)
{
if (sock_count == 0)
{
fprintf(stderr, "server(%d) - EOF from sock\n", port);
break;
}
else
{
perror2("server(%d) - read(sock)", port);
break;
}
}
DBG_MSG3("server(%d) - read %d bytes from sock", port, sock_count);
}
}
#if defined(__UNIX__)
/* Write to serial port possible? */
if (FD_ISSET(sio_fd, &wfds))
#elif defined(__WIN32__)
/* No socket IO performed? */
if ((!FD_ISSET(sock_fd, &rfds)) && (!FD_ISSET(sock_fd, &wfds)))
{
/* Break on a time out */
if (GetTickCount() - msecs > timeout)
{
fprintf(stderr, "server(%d) - timed out\n", port);
break;
}
}
else
{
msecs = GetTickCount();
}
if (1)
#endif
{
if (sock_count > 0)
{
if ((res = sio_write(&pipe->sio, sock_buf, sock_count)) < 0)
{
perror2("server(%d) - write(sio)", port);
break;
}
DBG_MSG3("server(%d) - wrote %d bytes to sio", port, res);
sock_count -= res;
}
}
}
/* Unlock our mutex */
thr_mutex_unlock(pipe->mutex);
}
fprintf(stderr, "server(%d) exiting\n", port);
/* Clean up - don't call pipe_cleanup() as that would nuke our mutex */
sio_cleanup(&pipe->sio);
tcp_cleanup(&pipe->sock);
free(pipe);
thr_exit((thr_exitcode_t)0);
return (thr_exitcode_t)0;
}
static int
bin_ztcp(char *nam, char **args, Options ops, UNUSED(int func))
{
int herrno, err=1, destport, force=0, verbose=0, test=0, targetfd=0;
ZSOCKLEN_T len;
char **addrp, *desthost;
const char *localname, *remotename;
struct hostent *zthost = NULL, *ztpeer = NULL;
struct servent *srv;
Tcp_session sess = NULL;
if (OPT_ISSET(ops,'f'))
force = 1;
if (OPT_ISSET(ops,'v'))
verbose = 1;
if (OPT_ISSET(ops,'t'))
test = 1;
if (OPT_ISSET(ops,'d')) {
targetfd = atoi(OPT_ARG(ops,'d'));
if (!targetfd) {
zwarnnam(nam, "%s is an invalid argument to -d", OPT_ARG(ops,'d'));
return 1;
}
}
if (OPT_ISSET(ops,'c')) {
if (!args[0]) {
tcp_cleanup();
}
else {
targetfd = atoi(args[0]);
sess = zts_byfd(targetfd);
if(!targetfd) {
zwarnnam(nam, "%s is an invalid argument to -c", args[0]);
return 1;
}
if (sess)
{
if ((sess->flags & ZTCP_ZFTP) && !force)
{
zwarnnam(nam, "use -f to force closure of a zftp control connection");
return 1;
}
tcp_close(sess);
return 0;
}
else
{
zwarnnam(nam, "fd %s not found in tcp table", args[0]);
return 1;
}
}
}
else if (OPT_ISSET(ops,'l')) {
int lport = 0;
if (!args[0]) {
zwarnnam(nam, "-l requires an argument");
return 1;
}
srv = getservbyname(args[0], "tcp");
if (srv)
lport = srv->s_port;
else
lport = htons(atoi(args[0]));
if (!lport) { zwarnnam(nam, "bad service name or port number");
return 1;
}
sess = tcp_socket(PF_INET, SOCK_STREAM, 0, ZTCP_LISTEN);
if (!sess) {
zwarnnam(nam, "unable to allocate a TCP session slot");
return 1;
}
#ifdef SO_OOBINLINE
len = 1;
setsockopt(sess->fd, SOL_SOCKET, SO_OOBINLINE, (char *)&len, sizeof(len));
#endif
if (!zsh_inet_aton("0.0.0.0", &(sess->sock.in.sin_addr)))
{
zwarnnam(nam, "bad address: %s", "0.0.0.0");
return 1;
}
sess->sock.in.sin_family = AF_INET;
sess->sock.in.sin_port = lport;
if (bind(sess->fd, (struct sockaddr *)&sess->sock.in, sizeof(struct sockaddr_in)))
{
char buf[DIGBUFSIZE];
convbase(buf, (zlong)ntohs(lport), 10);
zwarnnam(nam, "could not bind to port %s: %e", buf, errno);
tcp_close(sess);
return 1;
}
if (listen(sess->fd, 1))
{
zwarnnam(nam, "could not listen on socket: %e", errno);
tcp_close(sess);
return 1;
}
if (targetfd) {
sess->fd = redup(sess->fd, targetfd);
}
else {
/* move the fd since no one will want to read from it */
sess->fd = movefd(sess->fd);
}
if (sess->fd == -1) {
zwarnnam(nam, "cannot duplicate fd %d: %e", sess->fd, errno);
tcp_close(sess);
return 1;
}
setiparam_no_convert("REPLY", (zlong)sess->fd);
if (verbose)
printf("%d listener is on fd %d\n", ntohs(sess->sock.in.sin_port), sess->fd);
return 0;
}
else if (OPT_ISSET(ops,'a'))
{
int lfd, rfd;
if (!args[0]) {
zwarnnam(nam, "-a requires an argument");
return 1;
}
lfd = atoi(args[0]);
if (!lfd) {
zwarnnam(nam, "invalid numerical argument");
return 1;
}
sess = zts_byfd(lfd);
if (!sess) {
zwarnnam(nam, "fd %s is not registered as a tcp connection", args[0]);
return 1;
}
if (!(sess->flags & ZTCP_LISTEN))
{
zwarnnam(nam, "tcp connection not a listener");
return 1;
}
if (test) {
#if defined(HAVE_POLL) || defined(HAVE_SELECT)
# ifdef HAVE_POLL
struct pollfd pfd;
int ret;
pfd.fd = lfd;
pfd.events = POLLIN;
if ((ret = poll(&pfd, 1, 0)) == 0) return 1;
else if (ret == -1)
{
zwarnnam(nam, "poll error: %e", errno);
return 1;
}
# else
fd_set rfds;
struct timeval tv;
int ret;
FD_ZERO(&rfds);
FD_SET(lfd, &rfds);
tv.tv_sec = 0;
tv.tv_usec = 0;
if ((ret = select(lfd+1, &rfds, NULL, NULL, &tv)) == 0) return 1;
else if (ret == -1)
{
zwarnnam(nam, "select error: %e", errno);
return 1;
}
# endif
#else
zwarnnam(nam, "not currently supported");
return 1;
#endif
}
sess = zts_alloc(ZTCP_INBOUND);
len = sizeof(sess->peer.in);
do {
rfd = accept(lfd, (struct sockaddr *)&sess->peer.in, &len);
} while (rfd < 0 && errno == EINTR && !errflag);
if (rfd == -1) {
zwarnnam(nam, "could not accept connection: %e", errno);
tcp_close(sess);
return 1;
}
/* redup expects fd is already registered */
addmodulefd(rfd, FDT_MODULE);
if (targetfd) {
sess->fd = redup(rfd, targetfd);
if (sess->fd < 0) {
zerrnam(nam, "could not duplicate socket fd to %d: %e", targetfd, errno);
return 1;
}
}
else {
sess->fd = rfd;
}
setiparam_no_convert("REPLY", (zlong)sess->fd);
if (verbose)
printf("%d is on fd %d\n", ntohs(sess->peer.in.sin_port), sess->fd);
}
else
{
if (!args[0]) {
LinkNode node;
for(node = firstnode(ztcp_sessions); node; incnode(node))
{
sess = (Tcp_session)getdata(node);
if (sess->fd != -1)
{
zthost = gethostbyaddr((const void *)&(sess->sock.in.sin_addr), sizeof(sess->sock.in.sin_addr), AF_INET);
if (zthost)
localname = zthost->h_name;
else
localname = inet_ntoa(sess->sock.in.sin_addr);
ztpeer = gethostbyaddr((const void *)&(sess->peer.in.sin_addr), sizeof(sess->peer.in.sin_addr), AF_INET);
if (ztpeer)
remotename = ztpeer->h_name;
else
remotename = inet_ntoa(sess->peer.in.sin_addr);
if (OPT_ISSET(ops,'L')) {
int schar;
if (sess->flags & ZTCP_ZFTP)
schar = 'Z';
else if (sess->flags & ZTCP_LISTEN)
schar = 'L';
else if (sess->flags & ZTCP_INBOUND)
schar = 'I';
else
schar = 'O';
printf("%d %c %s %d %s %d\n",
sess->fd, schar,
localname, ntohs(sess->sock.in.sin_port),
remotename, ntohs(sess->peer.in.sin_port));
} else {
printf("%s:%d %s %s:%d is on fd %d%s\n",
localname, ntohs(sess->sock.in.sin_port),
((sess->flags & ZTCP_LISTEN) ? "-<" :
((sess->flags & ZTCP_INBOUND) ? "<-" : "->")),
remotename, ntohs(sess->peer.in.sin_port),
sess->fd,
(sess->flags & ZTCP_ZFTP) ? " ZFTP" : "");
}
}
}
return 0;
}
else if (!args[1]) {
destport = htons(23);
}
else {
srv = getservbyname(args[1],"tcp");
if (srv)
destport = srv->s_port;
else
destport = htons(atoi(args[1]));
}
desthost = ztrdup(args[0]);
zthost = zsh_getipnodebyname(desthost, AF_INET, 0, &herrno);
if (!zthost || errflag) {
zwarnnam(nam, "host resolution failure: %s", desthost);
zsfree(desthost);
return 1;
}
sess = tcp_socket(PF_INET, SOCK_STREAM, 0, 0);
if (!sess) {
zwarnnam(nam, "unable to allocate a TCP session slot");
zsfree(desthost);
return 1;
}
#ifdef SO_OOBINLINE
len = 1;
setsockopt(sess->fd, SOL_SOCKET, SO_OOBINLINE, (char *)&len, sizeof(len));
#endif
if (sess->fd < 0) {
zwarnnam(nam, "socket creation failed: %e", errno);
zsfree(desthost);
zts_delete(sess);
return 1;
}
for (addrp = zthost->h_addr_list; err && *addrp; addrp++) {
if (zthost->h_length != 4)
zwarnnam(nam, "address length mismatch");
do {
err = tcp_connect(sess, *addrp, zthost, destport);
} while (err && errno == EINTR && !errflag);
}
if (err) {
zwarnnam(nam, "connection failed: %e", errno);
tcp_close(sess);
zsfree(desthost);
return 1;
}
else
{
if (targetfd) {
sess->fd = redup(sess->fd, targetfd);
if (sess->fd < 0) {
zerrnam(nam, "could not duplicate socket fd to %d: %e", targetfd, errno);
zsfree(desthost);
tcp_close(sess);
return 1;
}
}
setiparam_no_convert("REPLY", (zlong)sess->fd);
if (verbose)
printf("%s:%d is now on fd %d\n",
desthost, destport, sess->fd);
}
zsfree(desthost);
}
return 0;
}
int main (int argc, char *argv[])
{
ASLogger *logger;
int stdin_handle;
input_id stdinput;
#ifdef WIN32
HANDLE hThread;
int fds[2];
#endif
/* winsock init */
tcp_startup ();
/* setup logging */
logger = as_logger_create ();
as_logger_add_output (logger, "stderr");
as_logger_add_output (logger, "ares.log");
AS_DBG ("Logging subsystem started");
/* setup event system */
as_event_init ();
/* init lib */
if (!as_init ())
{
printf ("FATA: as_init() failed\n");
exit (1);
}
#ifdef WIN32
/* create console reading thread on windows */
if (socketpair (0, 0, 0, fds) < 0)
{
printf ("FATAL: socketpair() failed\n");
exit (1);
}
stdin_handle = fds[1];
hThread = (HANDLE) _beginthreadex (NULL, 0, console_input_func,
(void *)fds[0], 0, NULL);
if (hThread == (HANDLE) -1 || hThread == (HANDLE) 0)
{
printf ("FATAL: couldn't start input thread\n");
exit (1);
}
#else
stdin_handle = 0;
#endif
/* add callback for command handling */
stdinput = input_add (stdin_handle, NULL, INPUT_READ, stdin_cb, 0);
#if 0
/* print prompt */
printf ("> ");
#endif
/* run event loop */
AS_DBG ("Entering event loop");
as_event_loop ();
AS_DBG ("Left event loop");
input_remove (stdinput);
#if WIN32
/* terminate thread if it is still running and close thread handle */
TerminateThread (hThread, 0);
CloseHandle (hThread);
#endif
/* cleanup lib */
as_cleanup ();
/* shutdown */
as_event_shutdown ();
as_logger_free (logger);
/* winsock shutdown */
tcp_cleanup ();
return 0;
}
