int main() {
int pid;
int fds[3];
int spair[2];
sys_chk(socketpair(AF_UNIX, SOCK_STREAM, 0, spair));
sys_chk(pid = fork());
if (pid == 0) { // Child.
sys_chk(fds[0] = open("/etc/passwd", O_RDONLY));
sys_chk(lseek(fds[0], 5, SEEK_SET));
sys_chk(fds[1] = open("/etc/group", O_RDONLY));
sys_chk(lseek(fds[1], 10, SEEK_SET));
sys_chk(fds[2] = open("/bin/sh", O_RDONLY));
sys_chk(lseek(fds[2], 20, SEEK_SET));
send_fds(spair[0], fds, 3);
return EXIT_SUCCESS;
} else { // Parent.
int *fds;
int fds_len;
recv_fds(spair[1], &fds, &fds_len);
printf("parent: got fds:");
for (int i = 0; i < fds_len; ++i) {
printf(" %d (fp=%lld)", fds[i], (long long)lseek(fds[i], 0, SEEK_CUR));
}
printf("\n");
return EXIT_SUCCESS;
}
}
static int send_criu_msg_with_fd(int socket_fd, CriuResp *msg, int fd)
{
unsigned char *buf;
int len, ret;
len = criu_resp__get_packed_size(msg);
buf = xmalloc(len);
if (!buf)
return -ENOMEM;
if (criu_resp__pack(msg, buf) != len) {
pr_perror("Failed packing response");
goto err;
}
if (fd >= 0) {
ret = send_fds(socket_fd, NULL, 0, &fd, 1, buf, len);
} else
ret = write(socket_fd, buf, len);
if (ret < 0) {
pr_perror("Can't send response");
goto err;
}
xfree(buf);
return 0;
err:
xfree(buf);
return -1;
}
static int drain_fds(struct parasite_drain_fd *args)
{
int ret;
ret = send_fds(tsock, NULL, 0,
args->fds, args->nr_fds, true);
if (ret)
pr_err("send_fds failed\n");
return ret;
}
int do_sendmsg(message *dev_m_in, message *dev_m_out)
{
int minor, peer, rc, i;
struct msg_control msg_ctrl;
#if DEBUG == 1
static int call_count = 0;
printf("(uds) [%d] do_sendmsg() call_count=%d\n",
uds_minor(dev_m_in), ++call_count);
#endif
minor = uds_minor(dev_m_in);
memset(&msg_ctrl, '\0', sizeof(struct msg_control));
rc = sys_safecopyfrom(VFS_PROC_NR, (cp_grant_id_t) dev_m_in->IO_GRANT,
(vir_bytes) 0, (vir_bytes) &msg_ctrl,
sizeof(struct msg_control));
if (rc != OK) {
return EIO;
}
/* locate peer */
peer = -1;
if (uds_fd_table[minor].type == SOCK_DGRAM) {
if (uds_fd_table[minor].target.sun_path[0] == '\0' ||
uds_fd_table[minor].target.sun_family != AF_UNIX) {
return EDESTADDRREQ;
}
for (i = 0; i < NR_FDS; i++) {
/* look for a SOCK_DGRAM socket that is bound on
* the target address
*/
if (uds_fd_table[i].type == SOCK_DGRAM &&
uds_fd_table[i].addr.sun_family == AF_UNIX &&
!strncmp(uds_fd_table[minor].target.sun_path,
uds_fd_table[i].addr.sun_path, UNIX_PATH_MAX)){
peer = i;
break;
}
}
if (peer == -1) {
return ENOENT;
}
} else {
peer = uds_fd_table[minor].peer;
if (peer == -1) {
return ENOTCONN;
}
}
#if DEBUG == 1
printf("(uds) [%d] sendmsg() -- peer=%d\n", minor, peer);
#endif
/* note: it's possible that there is already some file
* descriptors in ancillary_data if the peer didn't call
* recvmsg() yet. That's okay. The receiver will
* get the current file descriptors plus the new ones.
*/
rc = msg_control_read(&msg_ctrl, &uds_fd_table[peer].ancillary_data,
minor);
if (rc != OK) {
return rc;
}
return send_fds(minor, &uds_fd_table[peer].ancillary_data);
}
int main(int argc, char ** argv) {
int listen_port = -1;
char listen_port_str[8];
const char * ctrl_socket_path = NULL;
{
int opt;
while ((opt = getopt(argc, argv, "p:hu:")) != EOF) {
switch (opt) {
case 'p' :
listen_port = atoi(optarg);
break;
case 'h' :
fprintf(stderr, "%s [-p port] [-u socket-path]\n", argv[0]);
fprintf(stderr, "default: -p 9134\n");
exit(0);
case 'u' :
ctrl_socket_path = optarg;
break;
}
}
argc -= optind;
argv += optind;
}
if (listen_port == -1) {
listen_port = 9134;
}
sprintf(listen_port_str, "%d", listen_port);
typedef std::vector<fd_ctx> server_sockets_t;
server_sockets_t server_sockets;
peer_sockets_t peer_sockets;
fd_ctx ctrl_socket, ctrl_socket_conn;
bool ctrl_socket_mode_listen = false;
bool decay_mode = false;
ctrl_socket.fd = -1;
ctrl_socket_conn.fd = -1;
int sockets_inherited = 0;
int epoll = epoll_create(1024);
if (epoll < 0) {
VPERROR("epoll_create"); exit(1);
}
if (ctrl_socket_path) {
int s = socket(PF_UNIX, SOCK_SEQPACKET, 0);
if (s < 0) {
VPERROR("socket(AF_UNIX)");
exit(1);
}
struct sockaddr_un sun;
sun.sun_family = AF_UNIX;
strncpy(sun.sun_path, ctrl_socket_path, sizeof(sun.sun_path));
if (connect(s, (sockaddr *) &sun, sizeof(sun))) {
if (errno == ECONNREFUSED || errno == ENOENT) {
if (errno == ECONNREFUSED) {
if (unlink(ctrl_socket_path) < 0) {
fprintf(stderr, "unlink(%s): %s\n", ctrl_socket_path, strerror(errno));
exit(1);
}
}
ctrl_socket_listen(s, ctrl_socket_path);
ctrl_socket.fd = s;
poll_in(epoll, &ctrl_socket);
ctrl_socket_mode_listen = true;
} else {
fprintf(stderr, "connect(%s): %s\n", ctrl_socket_path, strerror(errno));
}
} else {
char buf[16];
ssize_t n = send(s, "unlisten", sizeof("unlisten") - 1, 0);
if (n < 0) {
VPERROR("sendmsg");
exit(1);
} else if (n == 0) {
fprintf(stderr, "unexpected EOF\n");
exit(1);
}
// blocking read
n = recv(s, buf, sizeof(buf), 0);
if (strncmp(buf, "unlistening", strlen("unlistening")) != 0) {
fprintf(stderr, "running server reported: ");
fwrite(buf, n, 1, stderr);
exit(1);
}
ctrl_socket_conn.fd = s;
poll_in(epoll, &ctrl_socket_conn);
}
}
{
struct addrinfo hints, * ai_res;
hints.ai_family = AF_UNSPEC;
hints.ai_socktype = SOCK_STREAM;
hints.ai_protocol = IPPROTO_TCP;
hints.ai_flags = AI_PASSIVE;
int r = getaddrinfo(NULL, listen_port_str, &hints, &ai_res);
if (r) {
fprintf(stderr, "getaddrinfo: %s\n", gai_strerror(r));
exit(1);
}
for (struct addrinfo * ai = ai_res; ai; ai = ai->ai_next) {
int s = socket(ai->ai_family, ai->ai_socktype, ai->ai_protocol);
if (s < 0) {
VPERROR("socket"); exit(1);
}
if (ai->ai_family == AF_INET6) {
int on = 1;
if (setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
(char *)&on, sizeof(on)) == -1) {
VPERROR("setsockopt(IPV6_ONLY)");
exit(1);
}
}
{
int on = 1;
if (setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (char *) &on, sizeof(on)) == -1) {
VPERROR("setsockopt(REUSEADDR)");
exit(1);
}
}
if (bind(s, ai->ai_addr, ai->ai_addrlen) < 0) {
VPERROR("bind"); exit(1);
}
if (listen(s, 50) < 0) {
VPERROR("listen"); exit(1);
}
fd_ctx c;
c.fd = s;
c.is_server = true;
c.protocol = ai->ai_protocol;
char * strp = c.buf;
int slen = sizeof(c.buf);
if (ai->ai_family == AF_INET6) {
*strp++ = '[';
slen -= 2;
}
get_ip_str(ai->ai_addr, strp, slen);
if (ai->ai_family == AF_INET6) {
strcat(c.buf, "]");
}
sprintf(c.buf + strlen(c.buf), ":%d", listen_port);
server_sockets.push_back(c);
}
freeaddrinfo(ai_res);
}
for (int i = 0; i < server_sockets.size(); ++i) {
poll_in(epoll, &server_sockets[i]);
}
epoll_event epoll_events[32];
const int epoll_max_events = 32;
fd_ctx fd_ctx_finder;
signal(SIGUSR1, sigusr1);
signal(SIGPIPE, SIG_IGN);
total_sockets = server_sockets.size();
time_t status_time = time(NULL);
while (total_sockets) {
if (unlikely(got_sigusr1)) {
// close listening sockets
for (int i = 0; i < server_sockets.size(); ++i) {
fprintf(stderr, "close server %s\n", server_sockets[i].buf);
if (epoll_ctl(epoll, EPOLL_CTL_DEL, server_sockets[i].fd, NULL) < 0) {
VPERROR("epoll_ctl");
}
close(server_sockets[i].fd);
--total_sockets;
}
got_sigusr1 = false;
}
if (unlikely(status_time + 5 < time(NULL))) {
fprintf(stderr, "%d connections, %d identified peers\n", total_connections - server_sockets.size(), peer_sockets.size());
status_time = time(NULL);
}
int ep_num = epoll_wait(epoll, epoll_events, epoll_max_events, 1000);
if (unlikely(ep_num < 0)) {
if (errno == EINTR) continue;
VPERROR("epoll_wait"); continue;
}
bool epoll_restart = false;
for (int epi = 0; epi < ep_num && ! epoll_restart; ++epi) {
fd_ctx * ctxp = (fd_ctx *) epoll_events[epi].data.ptr;
if (unlikely(ctxp == &ctrl_socket)) {
sockaddr_storage ss;
socklen_t sl = sizeof(ss);
int nsock = accept(ctxp->fd, (sockaddr *) &ss, &sl);
if (nsock < 0) {
VPERROR("accept"); continue;
}
epoll_event ev;
ev.events = EPOLLIN;
ev.data.ptr = (void *) &ctrl_socket_conn;
if (epoll_ctl(epoll, EPOLL_CTL_ADD, nsock, &ev) < 0) {
VPERROR("epoll_ctl");
close(nsock);
continue;
}
// we only ever accept one ctrl client
if (epoll_ctl(epoll, EPOLL_CTL_DEL, ctrl_socket.fd, NULL) < 0) {
VPERROR("epoll_ctl");
close(nsock);
continue;
}
ctrl_socket_conn.fd = nsock;
} else if (unlikely(ctxp == &ctrl_socket_conn)) {
if (ctrl_socket_mode_listen) {
char buf[1024];
int n = read(ctxp->fd, buf, sizeof(buf));
if (n < 0) {
if (errno == EINTR || errno == EAGAIN) continue;
VPERROR("read");
close(ctxp->fd);
poll_in(epoll, &ctrl_socket);
} else if (n == 0) {
close(ctxp->fd);
poll_in(epoll, &ctrl_socket);
} else {
if (strncmp(buf, "unlisten", sizeof("unlisten") - 1) == 0) {
for (int i = 0; i < server_sockets.size(); ++i) {
fprintf(stderr, "close server %s\n", server_sockets[i].buf);
if (epoll_ctl(epoll, EPOLL_CTL_DEL, server_sockets[i].fd, NULL) < 0) {
VPERROR("epoll_ctl");
}
close(server_sockets[i].fd);
--total_sockets;
}
if (write(ctrl_socket_conn.fd, "unlistening", sizeof("unlistening") - 1) < 0) {
VPERROR("write");
} else {
int nsent = 0;
do {
nsent = send_fds(ctrl_socket_conn.fd, epoll, peer_sockets.begin(), peer_sockets.end(), &peer_sockets);
if (nsent) {
fprintf(stderr, "bulk send: %d\n", nsent);
}
} while (nsent && ! peer_sockets.empty());
epoll_restart = true;
decay_mode = true;
}
}
}
} else {
msghdr msg;
iovec iov;
optional_buf<MAX_CONTROL_MESSAGE_CONTROL_SIZE, (MAX_CONTROL_MESSAGE_TOTAL_SIZE > FDCTX_BUFFER_SIZE)> control;
char * controlp = control.placeholder ?
ctxp->buf + MAX_CONTROL_MESSAGE_SIZE :
control.value;
optional_buf<MAX_CONTROL_MESSAGE_SIZE, (MAX_CONTROL_MESSAGE_SIZE > FDCTX_BUFFER_SIZE)> buf;
char * bufp = buf.placeholder ? ctxp->buf : control.value;
iov.iov_base = bufp;
iov.iov_len = MAX_CONTROL_MESSAGE_SIZE;
msg.msg_name = NULL;
msg.msg_namelen = 0;
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
msg.msg_control = (void *) controlp;
msg.msg_controllen = MAX_CONTROL_MESSAGE_CONTROL_SIZE;
msg.msg_flags = 0;
int n = recvmsg(ctxp->fd, &msg, 0);
if (n < 0) {
VPERROR("recvmsg");
} else if (n == 0) {
fprintf(stderr, "unexpected close\n");
close(ctxp->fd);
} else {
if (strncmp((const char *) iov.iov_base, "desc", std::min(4, n)) == 0) {
cmsghdr * cmp = CMSG_FIRSTHDR(&msg);
if (cmp->cmsg_level != SOL_SOCKET || cmp->cmsg_type != SCM_RIGHTS) {
fprintf(stderr, "malformed control message: wrong type\n");
exit(1);
}
int * uidp = (int *) ((char *) iov.iov_base + 4);
int * uidpend = (int *) ((char *) iov.iov_base + n);
int fd_count = 0;
for (; uidp < uidpend; ++uidp, ++fd_count) {
int fd = * ((int *) CMSG_DATA(cmp) + fd_count);
++sockets_inherited;
++total_sockets;
fd_ctx * cp = new fd_ctx;
cp->fd = fd;
cp->faf_uid = *uidp;
cp->is_server = false;
cp->protocol = IPPROTO_TCP;
cp->buf_len = 0;
epoll_event ev;
ev.events = EPOLLIN;
ev.data.ptr = (void *) cp;
if (epoll_ctl(epoll, EPOLL_CTL_ADD, cp->fd, &ev) < 0) {
VPERROR("epoll_ctl");
--total_sockets;
close(cp->fd);
delete cp;
}
if (cp->faf_uid != -1) {
peer_sockets.insert(cp);
}
}
} else if (strncmp((const char *) iov.iov_base, "exit", std::min(4, n)) == 0) {
close(ctxp->fd);
int s = socket(PF_UNIX, SOCK_SEQPACKET, 0);
if (s < 0) {
VPERROR("socket(PF_UNIX)");
} else {
ctrl_socket_listen(s, ctrl_socket_path);
ctrl_socket.fd = s;
poll_in(epoll, &ctrl_socket);
ctrl_socket_mode_listen = true;
}
fprintf(stderr, "%d sockets inherited from the dead\n", sockets_inherited);
}
}
}
} else if (unlikely(ctxp->is_server && ctxp->protocol == IPPROTO_TCP)) {
sockaddr_storage saddr;
socklen_t saddrlen = sizeof(saddr);
int nsock = accept(ctxp->fd, (sockaddr *) &saddr, &saddrlen);
if (nsock < 0) {
VPERROR("accept");
} else {
++total_sockets;
fd_ctx * cp = new fd_ctx;
cp->fd = nsock;
cp->faf_uid = -1;
cp->is_server = false;
cp->protocol = IPPROTO_TCP;
cp->buf_len = 0;
epoll_event ev;
ev.events = EPOLLIN;
ev.data.ptr = (void *) cp;
if (epoll_ctl(epoll, EPOLL_CTL_ADD, nsock, &ev) < 0) {
VPERROR("epoll_ctl");
--total_sockets;
close(nsock);
delete cp;
}
}
} else {
if (unlikely(decay_mode && ctxp->buf_len == 0)) {
fprintf(stderr, "single send\n");
send_fd(ctrl_socket_conn.fd, epoll, ctxp);
if (ctxp->faf_uid != -1) {
peer_sockets.erase(ctxp);
}
continue; // -> next epoll result
}
int n = read(ctxp->fd, ctxp->buf + ctxp->buf_len, PEER_CTX_BUF_SIZE - ctxp->buf_len);
if (unlikely(n < 0)) {
if (errno != ECONNRESET && errno != EAGAIN && errno != EINTR) {
VPERROR("read");
}
continue;
} else if (unlikely(n == 0)) {
close(ctxp->fd);
--total_sockets;
if (ctxp->faf_uid != -1) {
peer_sockets.erase(ctxp);
}
ctxp->remove_myself_from_peer_caches();
--ctxp->refcount;
if (ctxp->refcount == 0) {
delete ctxp;
} else {
ctxp->faf_uid = -1;
}
} else {
ctxp->buf_len += n;
char * buf_head = ctxp->buf;
bool postprocess = true;
while (buf_head < ctxp->buf + ctxp->buf_len) {
proxy_msg_header * h = (proxy_msg_header *) buf_head;
const int buf_len = ctxp->buf + ctxp->buf_len - buf_head;
const int in_msg_size = ntohl(h->size);
if (buf_len < 4) {
break;
}
if (unlikely(buf_len > PEER_CTX_BUF_SIZE)) {
// message to big
if (epoll_ctl(epoll, EPOLL_CTL_DEL, ctxp->fd, NULL) < 0) {
VPERROR("epoll_ctl");
}
close(ctxp->fd);
--total_sockets;
if (ctxp->faf_uid != -1) {
peer_sockets.erase(ctxp);
}
ctxp->remove_myself_from_peer_caches();
--ctxp->refcount;
if (ctxp->refcount == 0) {
delete ctxp;
} else {
ctxp->faf_uid = -1;
}
postprocess = false;
break;
}
if (in_msg_size + 4 > buf_len) {
break;
}
if (unlikely(ctxp->faf_uid == -1)) {
proxy_msg_header_set_uid * hu = (proxy_msg_header_set_uid *) h;
ctxp->faf_uid = ntohs(hu->uid);
peer_sockets.insert(ctxp);
buf_head += in_msg_size + 4;
continue; // -> next message from this fd_ctx
}
// in decay mode we always drop, because we expect our
// caches and refcounts to be inconsistent
// we can decay without bookkeeping if we never send any packets
// out (== we never expect a context to exists unless epoll still
// knows about it)
if (! decay_mode) {
int uid = ntohs(h->destuid);
fd_ctx * peer = ctxp->peers.find(uid);
if (unlikely(! peer)) {
fd_ctx_finder.faf_uid = uid;
peer_sockets_t::iterator iter = peer_sockets.find(&fd_ctx_finder);
if (iter != peer_sockets.end()) {
peer = *iter;
ctxp->peers.add(peer);
} else {
buf_head += in_msg_size + 4;
continue;
}
}
int in_port = ntohs(h->port);
proxy_msg_header_to_peer * hout = (proxy_msg_header_to_peer *) (buf_head + OUT_HEADER_OFFSET_ADJ);
hout->port = htons(in_port);
const int out_size = in_msg_size - OUT_HEADER_OFFSET_ADJ;
hout->size = htonl(out_size);
{
int n = write(peer->fd, (char *) hout, out_size + 4);
if (unlikely(n < 0)) {
if (errno != ECONNRESET && errno != EPIPE) {
VPERROR("write");
}
} else if (unlikely(n != out_size + 4)) {
fprintf(stderr, "short write (%d of %d\n", n, out_size + 4);
}
}
}
buf_head += in_msg_size + 4;
}
if (likely(postprocess)) {
int new_buflen = ctxp->buf + ctxp->buf_len - buf_head;
if (unlikely(new_buflen && ctxp->buf != buf_head)) {
for (char * p = ctxp->buf; buf_head < ctxp->buf + ctxp->buf_len; ++p, ++buf_head) {
*p = *buf_head;
}
}
ctxp->buf_len = new_buflen;
}
// we want to get rid of clients as soon as possible and
// dont wait for them to send the next message to trigger it
if (unlikely(decay_mode && ctxp->buf_len == 0)) {
send_fd(ctrl_socket_conn.fd, epoll, ctxp);
if (ctxp->faf_uid != -1) {
peer_sockets.erase(ctxp);
}
}
}
}
}
}
if (decay_mode && ctrl_socket_path) {
close(ctrl_socket.fd);
unlink(ctrl_socket_path);
if (write(ctrl_socket_conn.fd, "exit", strlen("exit")) < 0) {
VPERROR("send");
}
}
fprintf(stderr, "exit due to %d sockets left to serve\n", total_sockets);
exit(0);
}
int send_fd(int ctrlsock, int epoll, fd_ctx * ctxp) {
return send_fds(ctrlsock, epoll, &ctxp, &ctxp + 1, (dummy_erase_container<fd_ctx *> *) NULL);
}
