void service_anvil_global_init(void)
{
struct service_anvil_global *anvil;
anvil = i_new(struct service_anvil_global, 1);
if (pipe(anvil->status_fd) < 0)
i_fatal("pipe() failed: %m");
if (pipe(anvil->blocking_fd) < 0)
i_fatal("pipe() failed: %m");
if (pipe(anvil->nonblocking_fd) < 0)
i_fatal("pipe() failed: %m");
if (socketpair(AF_UNIX, SOCK_STREAM, 0, anvil->log_fdpass_fd) < 0)
i_fatal("socketpair() failed: %m");
fd_set_nonblock(anvil->status_fd[0], TRUE);
fd_set_nonblock(anvil->status_fd[1], TRUE);
fd_set_nonblock(anvil->nonblocking_fd[1], TRUE);
fd_close_on_exec(anvil->status_fd[0], TRUE);
fd_close_on_exec(anvil->status_fd[1], TRUE);
fd_close_on_exec(anvil->blocking_fd[0], TRUE);
fd_close_on_exec(anvil->blocking_fd[1], TRUE);
fd_close_on_exec(anvil->nonblocking_fd[0], TRUE);
fd_close_on_exec(anvil->nonblocking_fd[1], TRUE);
fd_close_on_exec(anvil->log_fdpass_fd[0], TRUE);
fd_close_on_exec(anvil->log_fdpass_fd[1], TRUE);
anvil->kills =
service_process_notify_init(anvil->nonblocking_fd[1],
service_process_write_anvil_kill);
service_anvil_global = anvil;
}
struct tube* tube_create(void)
{
struct tube* tube = (struct tube*)calloc(1, sizeof(*tube));
int sv[2];
if(!tube) {
int err = errno;
log_err("tube_create: out of memory");
errno = err;
return NULL;
}
tube->sr = -1;
tube->sw = -1;
if(socketpair(AF_UNIX, SOCK_STREAM, 0, sv) == -1) {
int err = errno;
log_err("socketpair: %s", strerror(errno));
free(tube);
errno = err;
return NULL;
}
tube->sr = sv[0];
tube->sw = sv[1];
if(!fd_set_nonblock(tube->sr) || !fd_set_nonblock(tube->sw)) {
int err = errno;
log_err("tube: cannot set nonblocking");
tube_delete(tube);
errno = err;
return NULL;
}
return tube;
}
int
pipe_cloexec_nonblock(int fd[2])
{
#ifdef WIN32
return _pipe(fd, 512, _O_BINARY);
#else
int ret;
#ifdef HAVE_PIPE2
ret = pipe2(fd, O_CLOEXEC|O_NONBLOCK);
if (ret >= 0 || errno != ENOSYS)
return ret;
#endif
ret = pipe(fd);
if (ret >= 0) {
fd_set_cloexec(fd[0], true);
fd_set_cloexec(fd[1], true);
fd_set_nonblock(fd[0]);
fd_set_nonblock(fd[1]);
}
return ret;
#endif
}
static void test_istream_unix_server(int fd)
{
struct istream *input;
const unsigned char *data;
size_t size;
input = i_stream_create_unix(fd, 1024);
/* 1) simple read */
test_server_read_nofd(input, 1);
write_one(fd);
/* 2) fd was sent but we won't get it */
test_server_read_nofd(input, 2);
/* we still shouldn't have the fd */
fd_set_nonblock(fd, TRUE);
i_stream_unix_set_read_fd(input);
test_assert(i_stream_read_data(input, &data, &size, 0) == 0);
test_assert(i_stream_unix_get_read_fd(input) == -1);
fd_set_nonblock(fd, FALSE);
write_one(fd);
/* 3) the previous fd should be lost now */
test_server_read_nofd(input, 3);
write_one(fd);
/* 4) we should get the fd now */
test_server_read_fd(input, send_fd2, 4);
write_one(fd);
/* 5) the previous fd shouldn't be returned anymore */
i_stream_unix_set_read_fd(input);
test_server_read_nofd(input, 5);
write_one(fd);
/* 6) with i_stream_unix_unset_read_fd() we shouldn't get fd anymore */
i_stream_unix_unset_read_fd(input);
test_server_read_nofd(input, 6);
write_one(fd);
/* 7-8) two fds were sent, but we'll get only the first one */
i_stream_unix_set_read_fd(input);
test_server_read_fd(input, send_fd, 7);
test_server_read_nofd(input, 8);
write_one(fd);
/* 9-10) two fds were sent, and we'll get them both */
i_stream_unix_set_read_fd(input);
test_server_read_fd(input, send_fd, 9);
i_stream_unix_set_read_fd(input);
test_server_read_fd(input, send_fd2, 10);
write_one(fd);
i_stream_destroy(&input);
i_close_fd(&fd);
}
int
accept_cloexec_nonblock(int fd, struct sockaddr *address,
size_t *address_length_r)
{
int ret;
socklen_t address_length = *address_length_r;
#ifdef HAVE_ACCEPT4
ret = accept4(fd, address, &address_length,
SOCK_CLOEXEC|SOCK_NONBLOCK);
if (ret >= 0 || errno != ENOSYS) {
if (ret >= 0)
*address_length_r = address_length;
return ret;
}
#endif
ret = accept(fd, address, &address_length);
if (ret >= 0) {
fd_set_cloexec(ret, true);
fd_set_nonblock(ret);
*address_length_r = address_length;
}
return ret;
}
int sock_comm_buffer_init(struct sock_conn *conn)
{
int optval;
socklen_t size = SOCK_COMM_BUF_SZ;
socklen_t optlen = sizeof(socklen_t);
optval = 1;
if (setsockopt(conn->sock_fd, IPPROTO_TCP, TCP_NODELAY,
&optval, sizeof optval))
SOCK_LOG_ERROR("setsockopt failed\n");
fd_set_nonblock(conn->sock_fd);
rbinit(&conn->inbuf, SOCK_COMM_BUF_SZ);
rbinit(&conn->outbuf, SOCK_COMM_BUF_SZ);
if (setsockopt(conn->sock_fd, SOL_SOCKET, SO_RCVBUF, &size, optlen))
SOCK_LOG_ERROR("setsockopt failed\n");
if (setsockopt(conn->sock_fd, SOL_SOCKET, SO_SNDBUF, &size, optlen))
SOCK_LOG_ERROR("setsockopt failed\n");
if (!getsockopt(conn->sock_fd, SOL_SOCKET, SO_RCVBUF, &size, &optlen))
SOCK_LOG_INFO("SO_RCVBUF: %d\n", size);
optlen = sizeof(socklen_t);
if (!getsockopt(conn->sock_fd, SOL_SOCKET, SO_SNDBUF, &size, &optlen))
SOCK_LOG_INFO("SO_SNDBUF: %d\n", size);
return 0;
}
int tube_write_msg(struct tube* tube, uint8_t* buf, uint32_t len,
int nonblock)
{
ssize_t r, d;
int fd = tube->sw;
/* test */
if(nonblock) {
r = write(fd, &len, sizeof(len));
if(r == -1) {
if(errno==EINTR || errno==EAGAIN)
return -1;
log_err("tube msg write failed: %s", strerror(errno));
return -1; /* can still continue, perhaps */
}
} else r = 0;
if(!fd_set_block(fd))
return 0;
/* write remainder */
d = r;
while(d != (ssize_t)sizeof(len)) {
if((r=write(fd, ((char*)&len)+d, sizeof(len)-d)) == -1) {
if(errno == EAGAIN)
continue; /* temporarily unavail: try again*/
log_err("tube msg write failed: %s", strerror(errno));
(void)fd_set_nonblock(fd);
return 0;
}
d += r;
}
d = 0;
while(d != (ssize_t)len) {
if((r=write(fd, buf+d, len-d)) == -1) {
if(errno == EAGAIN)
continue; /* temporarily unavail: try again*/
log_err("tube msg write failed: %s", strerror(errno));
(void)fd_set_nonblock(fd);
return 0;
}
d += r;
}
if(!fd_set_nonblock(fd))
return 0;
return 1;
}
int
dup_cloexec(int oldfd)
{
int newfd = dup(oldfd);
if (newfd >= 0)
fd_set_nonblock(newfd);
return newfd;
}
static struct dsync_ibc *
cmd_dsync_icb_stream_init(struct dsync_cmd_context *ctx,
const char *name, const char *temp_prefix)
{
if (ctx->input == NULL) {
fd_set_nonblock(ctx->fd_in, TRUE);
fd_set_nonblock(ctx->fd_out, TRUE);
ctx->input = i_stream_create_fd(ctx->fd_in, (size_t)-1, FALSE);
ctx->output = o_stream_create_fd(ctx->fd_out, (size_t)-1, FALSE);
}
i_stream_ref(ctx->input);
o_stream_ref(ctx->output);
if (ctx->rawlog_path != NULL) {
iostream_rawlog_create_path(ctx->rawlog_path,
&ctx->input, &ctx->output);
}
return dsync_ibc_init_stream(ctx->input, ctx->output,
name, temp_prefix);
}
int
create_local_accept_sock(const char *path, int* noproto)
{
#ifdef HAVE_SYS_UN_H
int s;
struct sockaddr_un usock;
verbose(VERB_ALGO, "creating unix socket %s", path);
#ifdef HAVE_STRUCT_SOCKADDR_UN_SUN_LEN
/* this member exists on BSDs, not Linux */
usock.sun_len = (socklen_t)sizeof(usock);
#endif
usock.sun_family = AF_LOCAL;
/* length is 92-108, 104 on FreeBSD */
(void)strlcpy(usock.sun_path, path, sizeof(usock.sun_path));
if ((s = socket(AF_LOCAL, SOCK_STREAM, 0)) == -1) {
log_err("Cannot create local socket %s (%s)",
path, strerror(errno));
return -1;
}
if (unlink(path) && errno != ENOENT) {
/* The socket already exists and cannot be removed */
log_err("Cannot remove old local socket %s (%s)",
path, strerror(errno));
return -1;
}
if (bind(s, (struct sockaddr *)&usock,
(socklen_t)sizeof(struct sockaddr_un)) == -1) {
log_err("Cannot bind local socket %s (%s)",
path, strerror(errno));
return -1;
}
if (!fd_set_nonblock(s)) {
log_err("Cannot set non-blocking mode");
return -1;
}
if (listen(s, TCP_BACKLOG) == -1) {
log_err("can't listen: %s", strerror(errno));
return -1;
}
(void)noproto; /*unused*/
return s;
#else
(void)path;
log_err("Local sockets are not supported");
*noproto = 1;
return -1;
#endif
}
void sock_set_sockopts(int sock)
{
int optval;
optval = 1;
sock_set_sockopt_reuseaddr(sock);
if (setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, &optval, sizeof(optval)))
SOCK_LOG_ERROR("setsockopt nodelay failed\n");
fd_set_nonblock(sock);
}
void sig_init() {
int pipe_fd[2];
struct sigaction act;
struct signal_spec_s *ss;
// in a last-ditch attempt to recover from fatal errors, we might re-run
// the initialization. otherwise, this condition is never true.
if (sig_wake_rd >= 0) close(sig_wake_rd);
if (sig_wake_wr >= 0) close(sig_wake_wr);
if (!sig_main_pid)
sig_main_pid= getpid();
// initialize with 0
memset((void*)signals, 0, sizeof(signals));
// Create pipe and set non-blocking
if (pipe(pipe_fd) < 0
|| fcntl(pipe_fd[0], F_SETFD, FD_CLOEXEC) < 0
|| fcntl(pipe_fd[1], F_SETFD, FD_CLOEXEC) < 0
|| !fd_set_nonblock(pipe_fd[0])
|| !fd_set_nonblock(pipe_fd[1])
) {
fatal(EXIT_IMPOSSIBLE_SCENARIO, "signal pipe setup: %s", strerror(errno));
}
log_trace("pipe => (%d, %d)", pipe_fd[0], pipe_fd[1]);
sig_wake_rd= pipe_fd[0];
sig_wake_wr= pipe_fd[1];
// set signal handlers
memset(&act, 0, sizeof(act));
for (ss= signal_spec; ss->signum != 0; ss++) {
act.sa_handler= ss->handler;
if (sigaction(ss->signum, &act, NULL))
fatal(EXIT_IMPOSSIBLE_SCENARIO, "signal handler setup: %s", strerror(errno));
}
// capture our signal mask
if (!sigprocmask(SIG_SETMASK, NULL, &sig_mask_orig) == 0)
perror("sigprocmask(all)");
}
static struct istream *mail_raw_create_stream
(struct mail_user *ruser, int fd, time_t *mtime_r, const char **sender)
{
struct istream *input, *input2, *input_list[2];
const unsigned char *data;
size_t i, size;
int ret, tz;
char *env_sender = NULL;
*mtime_r = (time_t)-1;
fd_set_nonblock(fd, FALSE);
input = i_stream_create_fd(fd, 4096, FALSE);
input->blocking = TRUE;
/* If input begins with a From-line, drop it */
ret = i_stream_read_data(input, &data, &size, 5);
if (ret > 0 && size >= 5 && memcmp(data, "From ", 5) == 0) {
/* skip until the first LF */
i_stream_skip(input, 5);
while ( i_stream_read_data(input, &data, &size, 0) > 0 ) {
for (i = 0; i < size; i++) {
if (data[i] == '\n')
break;
}
if (i != size) {
(void)mbox_from_parse(data, i, mtime_r, &tz, &env_sender);
i_stream_skip(input, i + 1);
break;
}
i_stream_skip(input, size);
}
}
if (env_sender != NULL && sender != NULL) {
*sender = t_strdup(env_sender);
}
i_free(env_sender);
if (input->v_offset == 0) {
input2 = input;
i_stream_ref(input2);
} else {
input2 = i_stream_create_limit(input, (uoff_t)-1);
}
i_stream_unref(&input);
input_list[0] = input2; input_list[1] = NULL;
input = i_stream_create_seekable(input_list, MAIL_MAX_MEMORY_BUFFER,
seekable_fd_callback, (void*)ruser);
i_stream_unref(&input2);
return input;
}
int
socketpair_cloexec_nonblock(int domain, int type, int protocol, int sv[2])
{
int ret;
#if defined(SOCK_CLOEXEC) && defined(SOCK_NONBLOCK)
ret = socketpair(domain, type | SOCK_CLOEXEC | SOCK_NONBLOCK, protocol,
sv);
if (ret >= 0 || errno != EINVAL)
return ret;
#endif
ret = socketpair(domain, type, protocol, sv);
if (ret >= 0) {
fd_set_cloexec(sv[0], true);
fd_set_nonblock(sv[0]);
fd_set_cloexec(sv[1], true);
fd_set_nonblock(sv[1]);
}
return ret;
}
static void
run_cmd(struct dsync_cmd_context *ctx, const char *const *args)
{
int fd_in[2], fd_out[2], fd_err[2];
if (pipe(fd_in) < 0 || pipe(fd_out) < 0 || pipe(fd_err) < 0)
i_fatal("pipe() failed: %m");
ctx->remote_pid = fork();
switch (ctx->remote_pid) {
case -1:
i_fatal("fork() failed: %m");
case 0:
/* child, which will execute the proxy server. stdin/stdout
goes to pipes which we'll pass to proxy client. */
if (dup2(fd_in[0], STDIN_FILENO) < 0 ||
dup2(fd_out[1], STDOUT_FILENO) < 0 ||
dup2(fd_err[1], STDERR_FILENO) < 0)
i_fatal("dup2() failed: %m");
i_close_fd(&fd_in[0]);
i_close_fd(&fd_in[1]);
i_close_fd(&fd_out[0]);
i_close_fd(&fd_out[1]);
i_close_fd(&fd_err[0]);
i_close_fd(&fd_err[1]);
execvp_const(args[0], args);
default:
/* parent */
break;
}
i_close_fd(&fd_in[0]);
i_close_fd(&fd_out[1]);
i_close_fd(&fd_err[1]);
ctx->fd_in = fd_out[0];
ctx->fd_out = fd_in[1];
ctx->fd_err = fd_err[0];
if (ctx->remote_user_prefix) {
const char *prefix =
t_strdup_printf("%s\n", ctx->ctx.cur_username);
if (write_full(ctx->fd_out, prefix, strlen(prefix)) < 0)
i_fatal("write(remote out) failed: %m");
}
fd_set_nonblock(ctx->fd_err, TRUE);
ctx->err_stream = i_stream_create_fd(ctx->fd_err, IO_BLOCK_SIZE, FALSE);
i_stream_set_return_partial_line(ctx->err_stream, TRUE);
}
/** find or else create proxy for this remote client */
static struct proxy*
find_create_proxy(struct sockaddr_storage* from, socklen_t from_len,
fd_set* rorig, int* max, struct proxy** proxies, int serv_ip6,
struct timeval* now, struct timeval* reuse_timeout)
{
struct proxy* p;
struct timeval t;
for(p = *proxies; p; p = p->next) {
if(sockaddr_cmp(from, from_len, &p->addr, p->addr_len)==0)
return p;
}
/* possibly: reuse lapsed entries */
for(p = *proxies; p; p = p->next) {
if(p->numwait > p->numsent || p->numsent > p->numreturn)
continue;
t = *now;
dl_tv_subtract(&t, &p->lastuse);
if(dl_tv_smaller(&t, reuse_timeout))
continue;
/* yes! */
verbose(1, "reuse existing entry");
memmove(&p->addr, from, from_len);
p->addr_len = from_len;
p->numreuse++;
return p;
}
/* create new */
p = (struct proxy*)calloc(1, sizeof(*p));
if(!p) fatal_exit("out of memory");
p->s = socket(serv_ip6?AF_INET6:AF_INET, SOCK_DGRAM, 0);
if(p->s == -1) {
#ifndef USE_WINSOCK
fatal_exit("socket: %s", strerror(errno));
#else
fatal_exit("socket: %s", wsa_strerror(WSAGetLastError()));
#endif
}
fd_set_nonblock(p->s);
memmove(&p->addr, from, from_len);
p->addr_len = from_len;
p->next = *proxies;
*proxies = p;
FD_SET(FD_SET_T p->s, rorig);
if(p->s+1 > *max)
*max = p->s+1;
return p;
}
int
socket_cloexec_nonblock(int domain, int type, int protocol)
{
int fd;
#if defined(SOCK_CLOEXEC) && defined(SOCK_NONBLOCK)
fd = socket(domain, type | SOCK_CLOEXEC | SOCK_NONBLOCK, protocol);
if (fd >= 0 || errno != EINVAL)
return fd;
#endif
fd = socket(domain, type, protocol);
if (fd >= 0) {
fd_set_cloexec(fd, true);
fd_set_nonblock(fd);
}
return fd;
}
int tcp_setsockopts(int fd)
{
struct linger lopt;
int rc = 0;
rc = fd_set_nonblock( fd, true );
/* Set SO_LINGER socket option */
memset( &lopt, 0, sizeof( struct linger ));
lopt.l_onoff = 1;
lopt.l_linger = 200; /* 2 seconds */
if ( setsockopt( fd, SOL_SOCKET, SO_LINGER,
(char*) &lopt, sizeof( struct linger )) == -1 )
{
rc--;
DEBUG(('I',1,"can't set SO_LINGER socket option on stdout: %s", strerror( errno )));
}
return ( rc == 0 );
}
int
create_udp_sock(int family, int socktype, struct sockaddr* addr,
socklen_t addrlen, int v6only, int* inuse, int* noproto,
int rcv, int snd)
{
int s;
#if defined(IPV6_USE_MIN_MTU)
int on=1;
#endif
#ifdef IPV6_MTU
int mtu = IPV6_MIN_MTU;
#endif
#if !defined(SO_RCVBUFFORCE) && !defined(SO_RCVBUF)
(void)rcv;
#endif
#if !defined(SO_SNDBUFFORCE) && !defined(SO_SNDBUF)
(void)snd;
#endif
#ifndef IPV6_V6ONLY
(void)v6only;
#endif
if((s = socket(family, socktype, 0)) == -1) {
*inuse = 0;
#ifndef USE_WINSOCK
if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
*noproto = 1;
return -1;
}
log_err("can't create socket: %s", strerror(errno));
#else
if(WSAGetLastError() == WSAEAFNOSUPPORT ||
WSAGetLastError() == WSAEPROTONOSUPPORT) {
*noproto = 1;
return -1;
}
log_err("can't create socket: %s",
wsa_strerror(WSAGetLastError()));
#endif
*noproto = 0;
return -1;
}
if(rcv) {
#ifdef SO_RCVBUF
int got;
socklen_t slen = (socklen_t)sizeof(got);
# ifdef SO_RCVBUFFORCE
/* Linux specific: try to use root permission to override
* system limits on rcvbuf. The limit is stored in
* /proc/sys/net/core/rmem_max or sysctl net.core.rmem_max */
if(setsockopt(s, SOL_SOCKET, SO_RCVBUFFORCE, (void*)&rcv,
(socklen_t)sizeof(rcv)) < 0) {
if(errno != EPERM) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., SO_RCVBUFFORCE, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., SO_RCVBUFFORCE, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
# endif /* SO_RCVBUFFORCE */
if(setsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&rcv,
(socklen_t)sizeof(rcv)) < 0) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., SO_RCVBUF, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., SO_RCVBUF, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
/* check if we got the right thing or if system
* reduced to some system max. Warn if so */
if(getsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&got,
&slen) >= 0 && got < rcv/2) {
log_warn("so-rcvbuf %u was not granted. "
"Got %u. To fix: start with "
"root permissions(linux) or sysctl "
"bigger net.core.rmem_max(linux) or "
"kern.ipc.maxsockbuf(bsd) values.",
(unsigned)rcv, (unsigned)got);
}
# ifdef SO_RCVBUFFORCE
}
# endif
#endif /* SO_RCVBUF */
}
/* first do RCVBUF as the receive buffer is more important */
if(snd) {
#ifdef SO_SNDBUF
int got;
socklen_t slen = (socklen_t)sizeof(got);
# ifdef SO_SNDBUFFORCE
/* Linux specific: try to use root permission to override
* system limits on sndbuf. The limit is stored in
* /proc/sys/net/core/wmem_max or sysctl net.core.wmem_max */
if(setsockopt(s, SOL_SOCKET, SO_SNDBUFFORCE, (void*)&snd,
(socklen_t)sizeof(snd)) < 0) {
if(errno != EPERM) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., SO_SNDBUFFORCE, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., SO_SNDBUFFORCE, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
# endif /* SO_SNDBUFFORCE */
if(setsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&snd,
(socklen_t)sizeof(snd)) < 0) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., SO_SNDBUF, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., SO_SNDBUF, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
/* check if we got the right thing or if system
* reduced to some system max. Warn if so */
if(getsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&got,
&slen) >= 0 && got < snd/2) {
log_warn("so-sndbuf %u was not granted. "
"Got %u. To fix: start with "
"root permissions(linux) or sysctl "
"bigger net.core.wmem_max(linux) or "
"kern.ipc.maxsockbuf(bsd) values.",
(unsigned)snd, (unsigned)got);
}
# ifdef SO_SNDBUFFORCE
}
# endif
#endif /* SO_SNDBUF */
}
if(family == AF_INET6) {
# if defined(IPV6_V6ONLY)
if(v6only) {
int val=(v6only==2)?0:1;
if (setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
(void*)&val, (socklen_t)sizeof(val)) < 0) {
#ifndef USE_WINSOCK
log_err("setsockopt(..., IPV6_V6ONLY"
", ...) failed: %s", strerror(errno));
close(s);
#else
log_err("setsockopt(..., IPV6_V6ONLY"
", ...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
#endif
*noproto = 0;
*inuse = 0;
return -1;
}
}
# endif
# if defined(IPV6_USE_MIN_MTU)
/*
* There is no fragmentation of IPv6 datagrams
* during forwarding in the network. Therefore
* we do not send UDP datagrams larger than
* the minimum IPv6 MTU of 1280 octets. The
* EDNS0 message length can be larger if the
* network stack supports IPV6_USE_MIN_MTU.
*/
if (setsockopt(s, IPPROTO_IPV6, IPV6_USE_MIN_MTU,
(void*)&on, (socklen_t)sizeof(on)) < 0) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., IPV6_USE_MIN_MTU, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., IPV6_USE_MIN_MTU, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
# elif defined(IPV6_MTU)
/*
* On Linux, to send no larger than 1280, the PMTUD is
* disabled by default for datagrams anyway, so we set
* the MTU to use.
*/
if (setsockopt(s, IPPROTO_IPV6, IPV6_MTU,
(void*)&mtu, (socklen_t)sizeof(mtu)) < 0) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., IPV6_MTU, ...) failed: %s",
strerror(errno));
close(s);
# else
log_err("setsockopt(..., IPV6_MTU, ...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
# endif /* IPv6 MTU */
} else if(family == AF_INET) {
# if defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
int action = IP_PMTUDISC_DONT;
if (setsockopt(s, IPPROTO_IP, IP_MTU_DISCOVER,
&action, (socklen_t)sizeof(action)) < 0) {
log_err("setsockopt(..., IP_MTU_DISCOVER, "
"IP_PMTUDISC_DONT...) failed: %s",
strerror(errno));
# ifndef USE_WINSOCK
close(s);
# else
closesocket(s);
# endif
return -1;
}
# elif defined(IP_DONTFRAG)
int off = 0;
if (setsockopt(s, IPPROTO_IP, IP_DONTFRAG,
&off, (socklen_t)sizeof(off)) < 0) {
log_err("setsockopt(..., IP_DONTFRAG, ...) failed: %s",
strerror(errno));
# ifndef USE_WINSOCK
close(s);
# else
closesocket(s);
# endif
return -1;
}
# endif /* IPv4 MTU */
}
if(bind(s, (struct sockaddr*)addr, addrlen) != 0) {
*noproto = 0;
#ifndef USE_WINSOCK
#ifdef EADDRINUSE
*inuse = (errno == EADDRINUSE);
/* detect freebsd jail with no ipv6 permission */
if(family==AF_INET6 && errno==EINVAL)
*noproto = 1;
else if(errno != EADDRINUSE) {
log_err("can't bind socket: %s", strerror(errno));
log_addr(0, "failed address",
(struct sockaddr_storage*)addr, addrlen);
}
#endif /* EADDRINUSE */
close(s);
#else /* USE_WINSOCK */
if(WSAGetLastError() != WSAEADDRINUSE &&
WSAGetLastError() != WSAEADDRNOTAVAIL) {
log_err("can't bind socket: %s",
wsa_strerror(WSAGetLastError()));
log_addr(0, "failed address",
(struct sockaddr_storage*)addr, addrlen);
}
closesocket(s);
#endif
return -1;
}
if(!fd_set_nonblock(s)) {
*noproto = 0;
*inuse = 0;
#ifndef USE_WINSOCK
close(s);
#else
closesocket(s);
#endif
return -1;
}
return s;
}
/** delayer main service routine */
static void
service(const char* bind_str, int bindport, const char* serv_str,
size_t memsize, int delay_msec)
{
struct sockaddr_storage bind_addr, srv_addr;
socklen_t bind_len, srv_len;
struct ringbuf* ring = ring_create(memsize);
struct timeval delay, reuse;
sldns_buffer* pkt;
int i, s, listen_s;
#ifndef S_SPLINT_S
delay.tv_sec = delay_msec / 1000;
delay.tv_usec = (delay_msec % 1000)*1000;
#endif
reuse = delay; /* reuse is max(4*delay, 1 second) */
dl_tv_add(&reuse, &delay);
dl_tv_add(&reuse, &delay);
dl_tv_add(&reuse, &delay);
if(reuse.tv_sec == 0)
reuse.tv_sec = 1;
if(!extstrtoaddr(serv_str, &srv_addr, &srv_len)) {
printf("cannot parse forward address: %s\n", serv_str);
exit(1);
}
pkt = sldns_buffer_new(65535);
if(!pkt)
fatal_exit("out of memory");
if( signal(SIGINT, delayer_sigh) == SIG_ERR ||
#ifdef SIGHUP
signal(SIGHUP, delayer_sigh) == SIG_ERR ||
#endif
#ifdef SIGQUIT
signal(SIGQUIT, delayer_sigh) == SIG_ERR ||
#endif
#ifdef SIGBREAK
signal(SIGBREAK, delayer_sigh) == SIG_ERR ||
#endif
#ifdef SIGALRM
signal(SIGALRM, delayer_sigh) == SIG_ERR ||
#endif
signal(SIGTERM, delayer_sigh) == SIG_ERR)
fatal_exit("could not bind to signal");
/* bind UDP port */
if((s = socket(str_is_ip6(bind_str)?AF_INET6:AF_INET,
SOCK_DGRAM, 0)) == -1) {
#ifndef USE_WINSOCK
fatal_exit("socket: %s", strerror(errno));
#else
fatal_exit("socket: %s", wsa_strerror(WSAGetLastError()));
#endif
}
i=0;
if(bindport == 0) {
bindport = 1024 + random()%64000;
i = 100;
}
while(1) {
if(!ipstrtoaddr(bind_str, bindport, &bind_addr, &bind_len)) {
printf("cannot parse listen address: %s\n", bind_str);
exit(1);
}
if(bind(s, (struct sockaddr*)&bind_addr, bind_len) == -1) {
#ifndef USE_WINSOCK
log_err("bind: %s", strerror(errno));
#else
log_err("bind: %s", wsa_strerror(WSAGetLastError()));
#endif
if(i--==0)
fatal_exit("cannot bind any port");
bindport = 1024 + random()%64000;
} else break;
}
fd_set_nonblock(s);
/* and TCP port */
if((listen_s = socket(str_is_ip6(bind_str)?AF_INET6:AF_INET,
SOCK_STREAM, 0)) == -1) {
#ifndef USE_WINSOCK
fatal_exit("tcp socket: %s", strerror(errno));
#else
fatal_exit("tcp socket: %s", wsa_strerror(WSAGetLastError()));
#endif
}
#ifdef SO_REUSEADDR
if(1) {
int on = 1;
if(setsockopt(listen_s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
(socklen_t)sizeof(on)) < 0)
#ifndef USE_WINSOCK
fatal_exit("setsockopt(.. SO_REUSEADDR ..) failed: %s",
strerror(errno));
#else
fatal_exit("setsockopt(.. SO_REUSEADDR ..) failed: %s",
wsa_strerror(WSAGetLastError()));
#endif
}
#endif
if(bind(listen_s, (struct sockaddr*)&bind_addr, bind_len) == -1) {
#ifndef USE_WINSOCK
fatal_exit("tcp bind: %s", strerror(errno));
#else
fatal_exit("tcp bind: %s", wsa_strerror(WSAGetLastError()));
#endif
}
if(listen(listen_s, 5) == -1) {
#ifndef USE_WINSOCK
fatal_exit("tcp listen: %s", strerror(errno));
#else
fatal_exit("tcp listen: %s", wsa_strerror(WSAGetLastError()));
#endif
}
fd_set_nonblock(listen_s);
printf("listening on port: %d\n", bindport);
/* process loop */
do_quit = 0;
service_loop(s, listen_s, ring, &delay, &reuse, &srv_addr, srv_len,
pkt);
/* cleanup */
verbose(1, "cleanup");
#ifndef USE_WINSOCK
close(s);
close(listen_s);
#else
closesocket(s);
closesocket(listen_s);
#endif
sldns_buffer_free(pkt);
ring_delete(ring);
}
struct sock_conn *sock_ep_connect(struct sock_ep *ep, fi_addr_t index)
{
int conn_fd = -1, ret;
int do_retry = sock_conn_retry;
struct sock_conn *conn, *new_conn;
uint16_t idx;
struct sockaddr_in *addr;
socklen_t lon;
int valopt = 0;
struct pollfd poll_fd;
if (ep->ep_type == FI_EP_MSG) {
idx = 0;
addr = ep->dest_addr;
} else {
idx = index & ep->av->mask;
addr = (struct sockaddr_in *)&ep->av->table[idx].addr;
}
do_connect:
fastlock_acquire(&ep->cmap.lock);
conn = sock_ep_lookup_conn(ep, index, addr);
fastlock_release(&ep->cmap.lock);
if (conn != SOCK_CM_CONN_IN_PROGRESS)
return conn;
conn_fd = socket(AF_INET, SOCK_STREAM, 0);
if (conn_fd == -1) {
SOCK_LOG_ERROR("failed to create conn_fd, errno: %d\n", errno);
errno = FI_EOTHER;
return NULL;
}
ret = fd_set_nonblock(conn_fd);
if (ret) {
SOCK_LOG_ERROR("failed to set conn_fd nonblocking, errno: %d\n", errno);
errno = FI_EOTHER;
close(conn_fd);
return NULL;
}
SOCK_LOG_DBG("Connecting to: %s:%d\n", inet_ntoa(addr->sin_addr),
ntohs(addr->sin_port));
SOCK_LOG_DBG("Connecting using address:%s\n",
inet_ntoa(ep->src_addr->sin_addr));
ret = connect(conn_fd, (struct sockaddr *) addr, sizeof *addr);
if (ret < 0) {
if (errno == EINPROGRESS) {
poll_fd.fd = conn_fd;
poll_fd.events = POLLOUT;
ret = poll(&poll_fd, 1, 15 * 1000);
if (ret < 0) {
SOCK_LOG_DBG("poll failed\n");
goto retry;
}
lon = sizeof(int);
ret = getsockopt(conn_fd, SOL_SOCKET, SO_ERROR, (void*)(&valopt), &lon);
if (ret < 0) {
SOCK_LOG_DBG("getsockopt failed: %d, %d\n", ret, conn_fd);
goto retry;
}
if (valopt) {
SOCK_LOG_DBG("Error in connection() %d - %s - %d\n", valopt, strerror(valopt), conn_fd);
SOCK_LOG_DBG("Connecting to: %s:%d\n", inet_ntoa(addr->sin_addr),
ntohs(addr->sin_port));
SOCK_LOG_DBG("Connecting using address:%s\n",
inet_ntoa(ep->src_addr->sin_addr));
goto retry;
}
goto out;
} else {
SOCK_LOG_DBG("Timeout or error() - %s: %d\n", strerror(errno), conn_fd);
SOCK_LOG_DBG("Connecting to: %s:%d\n", inet_ntoa(addr->sin_addr),
ntohs(addr->sin_port));
SOCK_LOG_DBG("Connecting using address:%s\n",
inet_ntoa(ep->src_addr->sin_addr));
goto retry;
}
} else {
goto out;
}
retry:
do_retry--;
sleep(10);
if (!do_retry)
goto err;
if (conn_fd != -1) {
close(conn_fd);
conn_fd = -1;
}
SOCK_LOG_ERROR("Connect error, retrying - %s - %d\n", strerror(errno), conn_fd);
SOCK_LOG_DBG("Connecting to: %s:%d\n", inet_ntoa(addr->sin_addr),
ntohs(addr->sin_port));
SOCK_LOG_DBG("Connecting using address:%s\n",
inet_ntoa(ep->src_addr->sin_addr));
goto do_connect;
out:
fastlock_acquire(&ep->cmap.lock);
new_conn = sock_conn_map_insert(ep, addr, conn_fd, 0);
new_conn->av_index = (ep->ep_type == FI_EP_MSG) ? FI_ADDR_NOTAVAIL : (fi_addr_t) idx;
conn = idm_lookup(&ep->av_idm, index);
if (conn == SOCK_CM_CONN_IN_PROGRESS) {
idm_set(&ep->av_idm, index, new_conn);
conn = new_conn;
}
fastlock_release(&ep->cmap.lock);
return conn;
err:
close(conn_fd);
return NULL;
}
static
void test_iostream_proxy_simple(void)
{
size_t bytes;
test_begin("iostream_proxy");
int sfdl[2];
int sfdr[2];
int counter;
test_assert(socketpair(AF_UNIX, SOCK_STREAM, 0, sfdl) == 0);
test_assert(socketpair(AF_UNIX, SOCK_STREAM, 0, sfdr) == 0);
fd_set_nonblock(sfdl[0], TRUE);
fd_set_nonblock(sfdl[1], TRUE);
fd_set_nonblock(sfdr[0], TRUE);
fd_set_nonblock(sfdr[1], TRUE);
struct ioloop *ioloop = io_loop_create();
struct istream *left_in = i_stream_create_fd(sfdl[1], IO_BLOCK_SIZE);
struct ostream *left_out = o_stream_create_fd(sfdl[1], IO_BLOCK_SIZE);
struct istream *right_in = i_stream_create_fd(sfdr[1], IO_BLOCK_SIZE);
struct ostream *right_out = o_stream_create_fd(sfdr[1], IO_BLOCK_SIZE);
struct iostream_proxy *proxy;
proxy = iostream_proxy_create(left_in, left_out, right_in, right_out);
i_stream_unref(&left_in);
o_stream_unref(&left_out);
i_stream_unref(&right_in);
o_stream_unref(&right_out);
iostream_proxy_set_completion_callback(proxy, completed, &counter);
iostream_proxy_start(proxy);
left_in = i_stream_create_fd(sfdl[0], IO_BLOCK_SIZE);
left_out = o_stream_create_fd(sfdl[0], IO_BLOCK_SIZE);
right_in = i_stream_create_fd(sfdr[0], IO_BLOCK_SIZE);
right_out = o_stream_create_fd(sfdr[0], IO_BLOCK_SIZE);
test_assert(proxy != NULL);
test_assert(o_stream_send_str(left_out, "hello, world") > 0);
o_stream_flush(left_out);
o_stream_unref(&left_out);
test_assert(shutdown(sfdl[0], SHUT_WR) == 0);
counter = 1;
io_loop_run(ioloop);
i_stream_read(right_in);
test_assert(strcmp((const char*)i_stream_get_data(right_in, &bytes), "hello, world") == 0);
i_stream_skip(right_in, bytes);
test_assert(o_stream_send_str(right_out, "hello, world") > 0);
o_stream_flush(right_out);
o_stream_unref(&right_out);
test_assert(shutdown(sfdr[0], SHUT_WR) == 0);
counter = 1;
io_loop_run(ioloop);
i_stream_read(left_in);
test_assert(strcmp((const char*)i_stream_get_data(left_in, &bytes), "hello, world") == 0);
i_stream_skip(left_in, bytes);
iostream_proxy_unref(&proxy);
io_loop_destroy(&ioloop);
i_stream_unref(&left_in);
i_stream_unref(&right_in);
/* close fd */
close(sfdl[0]);
close(sfdl[1]);
close(sfdr[0]);
close(sfdr[1]);
test_end();
}
static int __backlight_init(struct backlight *b, char *iface, int fd)
{
b->fd = fd_move_cloexec(fd_set_nonblock(fd));
b->iface = iface;
return 1;
}
struct sock_conn *sock_ep_connect(struct sock_ep_attr *ep_attr, fi_addr_t index)
{
int conn_fd = -1, ret;
int do_retry = sock_conn_retry;
struct sock_conn *conn, *new_conn;
struct sockaddr_in addr;
socklen_t lon;
int valopt = 0;
struct pollfd poll_fd;
if (ep_attr->ep_type == FI_EP_MSG) {
/* Need to check that destination address has been
passed to endpoint */
assert(ep_attr->dest_addr);
addr = *ep_attr->dest_addr;
addr.sin_port = htons(ep_attr->msg_dest_port);
} else {
addr = *((struct sockaddr_in *)&ep_attr->av->table[index].addr);
}
do_connect:
fastlock_acquire(&ep_attr->cmap.lock);
conn = sock_ep_lookup_conn(ep_attr, index, &addr);
fastlock_release(&ep_attr->cmap.lock);
if (conn != SOCK_CM_CONN_IN_PROGRESS)
return conn;
conn_fd = ofi_socket(AF_INET, SOCK_STREAM, 0);
if (conn_fd == -1) {
SOCK_LOG_ERROR("failed to create conn_fd, errno: %d\n", errno);
errno = FI_EOTHER;
return NULL;
}
ret = fd_set_nonblock(conn_fd);
if (ret) {
SOCK_LOG_ERROR("failed to set conn_fd nonblocking, errno: %d\n", errno);
errno = FI_EOTHER;
ofi_close_socket(conn_fd);
return NULL;
}
SOCK_LOG_DBG("Connecting to: %s:%d\n", inet_ntoa(addr.sin_addr),
ntohs(addr.sin_port));
SOCK_LOG_DBG("Connecting using address:%s\n",
inet_ntoa(ep_attr->src_addr->sin_addr));
ret = connect(conn_fd, (struct sockaddr *) &addr, sizeof addr);
if (ret < 0) {
if (ofi_sockerr() == EINPROGRESS) {
poll_fd.fd = conn_fd;
poll_fd.events = POLLOUT;
ret = poll(&poll_fd, 1, 15 * 1000);
if (ret < 0) {
SOCK_LOG_DBG("poll failed\n");
goto retry;
}
lon = sizeof(int);
ret = getsockopt(conn_fd, SOL_SOCKET, SO_ERROR, (void*)(&valopt), &lon);
if (ret < 0) {
SOCK_LOG_DBG("getsockopt failed: %d, %d\n", ret, conn_fd);
goto retry;
}
if (valopt) {
SOCK_LOG_DBG("Error in connection() %d - %s - %d\n", valopt, strerror(valopt), conn_fd);
SOCK_LOG_DBG("Connecting to: %s:%d\n", inet_ntoa(addr.sin_addr),
ntohs(addr.sin_port));
SOCK_LOG_DBG("Connecting using address:%s\n",
inet_ntoa(ep_attr->src_addr->sin_addr));
goto retry;
}
goto out;
} else {
SOCK_LOG_DBG("Timeout or error() - %s: %d\n", strerror(errno), conn_fd);
SOCK_LOG_DBG("Connecting to: %s:%d\n", inet_ntoa(addr.sin_addr),
ntohs(addr.sin_port));
SOCK_LOG_DBG("Connecting using address:%s\n",
inet_ntoa(ep_attr->src_addr->sin_addr));
goto retry;
}
} else {
goto out;
}
retry:
do_retry--;
sleep(10);
if (!do_retry)
goto err;
if (conn_fd != -1) {
ofi_close_socket(conn_fd);
conn_fd = -1;
}
SOCK_LOG_ERROR("Connect error, retrying - %s - %d\n", strerror(errno), conn_fd);
SOCK_LOG_DBG("Connecting to: %s:%d\n", inet_ntoa(addr.sin_addr),
ntohs(addr.sin_port));
SOCK_LOG_DBG("Connecting using address:%s\n",
inet_ntoa(ep_attr->src_addr->sin_addr));
goto do_connect;
out:
fastlock_acquire(&ep_attr->cmap.lock);
new_conn = sock_conn_map_insert(ep_attr, &addr, conn_fd, 0);
if (!new_conn) {
fastlock_release(&ep_attr->cmap.lock);
goto err;
}
new_conn->av_index = (ep_attr->ep_type == FI_EP_MSG) ? FI_ADDR_NOTAVAIL : index;
conn = ofi_idm_lookup(&ep_attr->av_idm, index);
if (conn == SOCK_CM_CONN_IN_PROGRESS) {
if (ofi_idm_set(&ep_attr->av_idm, index, new_conn) < 0)
SOCK_LOG_ERROR("ofi_idm_set failed\n");
conn = new_conn;
}
fastlock_release(&ep_attr->cmap.lock);
return conn;
err:
ofi_close_socket(conn_fd);
return NULL;
}
int
create_tcp_accept_sock(struct addrinfo *addr, int v6only, int* noproto,
int* reuseport)
{
int s;
#if defined(SO_REUSEADDR) || defined(SO_REUSEPORT) || defined(IPV6_V6ONLY)
int on = 1;
#endif /* SO_REUSEADDR || IPV6_V6ONLY */
verbose_print_addr(addr);
*noproto = 0;
if((s = socket(addr->ai_family, addr->ai_socktype, 0)) == -1) {
#ifndef USE_WINSOCK
if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
*noproto = 1;
return -1;
}
log_err("can't create socket: %s", strerror(errno));
#else
if(WSAGetLastError() == WSAEAFNOSUPPORT ||
WSAGetLastError() == WSAEPROTONOSUPPORT) {
*noproto = 1;
return -1;
}
log_err("can't create socket: %s",
wsa_strerror(WSAGetLastError()));
#endif
return -1;
}
#ifdef SO_REUSEADDR
if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
(socklen_t)sizeof(on)) < 0) {
#ifndef USE_WINSOCK
log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
strerror(errno));
close(s);
#else
log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
#endif
return -1;
}
#endif /* SO_REUSEADDR */
#ifdef SO_REUSEPORT
/* try to set SO_REUSEPORT so that incoming
* connections are distributed evenly among the receiving threads.
* Each thread must have its own socket bound to the same port,
* with SO_REUSEPORT set on each socket.
*/
if (reuseport && *reuseport &&
setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void*)&on,
(socklen_t)sizeof(on)) < 0) {
#ifdef ENOPROTOOPT
if(errno != ENOPROTOOPT || verbosity >= 3)
log_warn("setsockopt(.. SO_REUSEPORT ..) failed: %s",
strerror(errno));
#endif
/* this option is not essential, we can continue */
*reuseport = 0;
}
#else
(void)reuseport;
#endif /* defined(SO_REUSEPORT) */
#if defined(IPV6_V6ONLY)
if(addr->ai_family == AF_INET6 && v6only) {
if(setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
(void*)&on, (socklen_t)sizeof(on)) < 0) {
#ifndef USE_WINSOCK
log_err("setsockopt(..., IPV6_V6ONLY, ...) failed: %s",
strerror(errno));
close(s);
#else
log_err("setsockopt(..., IPV6_V6ONLY, ...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
#endif
return -1;
}
}
#else
(void)v6only;
#endif /* IPV6_V6ONLY */
if(bind(s, addr->ai_addr, addr->ai_addrlen) != 0) {
#ifndef USE_WINSOCK
/* detect freebsd jail with no ipv6 permission */
if(addr->ai_family==AF_INET6 && errno==EINVAL)
*noproto = 1;
else {
log_err_addr("can't bind socket", strerror(errno),
(struct sockaddr_storage*)addr->ai_addr,
addr->ai_addrlen);
}
close(s);
#else
log_err_addr("can't bind socket",
wsa_strerror(WSAGetLastError()),
(struct sockaddr_storage*)addr->ai_addr,
addr->ai_addrlen);
closesocket(s);
#endif
return -1;
}
if(!fd_set_nonblock(s)) {
#ifndef USE_WINSOCK
close(s);
#else
closesocket(s);
#endif
return -1;
}
if(listen(s, TCP_BACKLOG) == -1) {
#ifndef USE_WINSOCK
log_err("can't listen: %s", strerror(errno));
close(s);
#else
log_err("can't listen: %s", wsa_strerror(WSAGetLastError()));
closesocket(s);
#endif
return -1;
}
return s;
}
int sock_conn_listen(struct sock_ep_attr *ep_attr)
{
struct addrinfo *s_res = NULL, *p;
struct addrinfo hints = { 0 };
int listen_fd = 0, ret;
socklen_t addr_size;
struct sockaddr_in addr;
struct sock_conn_listener *listener = &ep_attr->listener;
char service[NI_MAXSERV] = {0};
char *port;
char ipaddr[24];
hints.ai_family = AF_INET;
hints.ai_socktype = SOCK_STREAM;
hints.ai_flags = AI_PASSIVE;
memcpy(&addr, ep_attr->src_addr, sizeof(addr));
if (getnameinfo((void *)ep_attr->src_addr, sizeof(*ep_attr->src_addr),
NULL, 0, listener->service,
sizeof(listener->service), NI_NUMERICSERV)) {
SOCK_LOG_ERROR("could not resolve src_addr\n");
return -FI_EINVAL;
}
if (ep_attr->ep_type == FI_EP_MSG) {
memset(listener->service, 0, NI_MAXSERV);
port = NULL;
addr.sin_port = 0;
} else
port = listener->service;
inet_ntop(addr.sin_family, &addr.sin_addr, ipaddr, sizeof(ipaddr));
ret = getaddrinfo(ipaddr, port, &hints, &s_res);
if (ret) {
SOCK_LOG_ERROR("no available AF_INET address, service %s, %s\n",
listener->service, gai_strerror(ret));
return -FI_EINVAL;
}
SOCK_LOG_DBG("Binding listener thread to port: %s\n", listener->service);
for (p = s_res; p; p = p->ai_next) {
listen_fd = ofi_socket(p->ai_family, p->ai_socktype, p->ai_protocol);
if (listen_fd >= 0) {
sock_set_sockopts(listen_fd);
if (!bind(listen_fd, s_res->ai_addr, s_res->ai_addrlen))
break;
ofi_close_socket(listen_fd);
listen_fd = -1;
}
}
freeaddrinfo(s_res);
if (listen_fd < 0) {
SOCK_LOG_ERROR("failed to listen to port: %s\n",
listener->service);
goto err;
}
if (atoi(listener->service) == 0) {
addr_size = sizeof(addr);
if (getsockname(listen_fd, (struct sockaddr *) &addr, &addr_size))
goto err;
snprintf(listener->service, sizeof listener->service, "%d",
ntohs(addr.sin_port));
SOCK_LOG_DBG("Bound to port: %s - %d\n", listener->service, getpid());
ep_attr->msg_src_port = ntohs(addr.sin_port);
}
if (ep_attr->src_addr->sin_addr.s_addr == 0) {
snprintf(service, sizeof service, "%s", listener->service);
ret = sock_get_src_addr_from_hostname(ep_attr->src_addr, service);
if (ret)
goto err;
}
if (listen(listen_fd, sock_cm_def_map_sz)) {
SOCK_LOG_ERROR("failed to listen socket: %s\n", strerror(errno));
goto err;
}
if (((struct sockaddr_in *) (ep_attr->src_addr))->sin_port == 0) {
((struct sockaddr_in *) (ep_attr->src_addr))->sin_port =
htons(atoi(listener->service));
}
listener->sock = listen_fd;
if (socketpair(AF_UNIX, SOCK_STREAM, 0, listener->signal_fds) < 0)
goto err;
listener->do_listen = 1;
fd_set_nonblock(listener->signal_fds[1]);
if (pthread_create(&listener->listener_thread, 0,
_sock_conn_listen, ep_attr)) {
SOCK_LOG_ERROR("failed to create conn listener thread\n");
goto err;
} while (!*((volatile int*)&listener->is_ready));
return 0;
err:
if (listen_fd >= 0)
ofi_close_socket(listen_fd);
return -FI_EINVAL;
}
int tube_read_msg(struct tube* tube, uint8_t** buf, uint32_t* len,
int nonblock)
{
ssize_t r, d;
int fd = tube->sr;
/* test */
*len = 0;
if(nonblock) {
r = read(fd, len, sizeof(*len));
if(r == -1) {
if(errno==EINTR || errno==EAGAIN)
return -1;
log_err("tube msg read failed: %s", strerror(errno));
return -1; /* we can still continue, perhaps */
}
if(r == 0) /* EOF */
return 0;
} else r = 0;
if(!fd_set_block(fd))
return 0;
/* read remainder */
d = r;
while(d != (ssize_t)sizeof(*len)) {
if((r=read(fd, ((char*)len)+d, sizeof(*len)-d)) == -1) {
log_err("tube msg read failed: %s", strerror(errno));
(void)fd_set_nonblock(fd);
return 0;
}
if(r == 0) /* EOF */ {
(void)fd_set_nonblock(fd);
return 0;
}
d += r;
}
log_assert(*len < 65536*2);
*buf = (uint8_t*)malloc(*len);
if(!*buf) {
log_err("tube read out of memory");
(void)fd_set_nonblock(fd);
return 0;
}
d = 0;
while(d < (ssize_t)*len) {
if((r=read(fd, (*buf)+d, (size_t)((ssize_t)*len)-d)) == -1) {
log_err("tube msg read failed: %s", strerror(errno));
(void)fd_set_nonblock(fd);
free(*buf);
return 0;
}
if(r == 0) { /* EOF */
(void)fd_set_nonblock(fd);
free(*buf);
return 0;
}
d += r;
}
if(!fd_set_nonblock(fd)) {
free(*buf);
return 0;
}
return 1;
}
/** accept new TCP connections, and set them up */
static void
service_tcp_listen(int s, fd_set* rorig, int* max, struct tcp_proxy** proxies,
struct sockaddr_storage* srv_addr, socklen_t srv_len,
struct timeval* now, struct timeval* tcp_timeout)
{
int newfd;
struct sockaddr_storage addr;
struct tcp_proxy* p;
socklen_t addr_len;
newfd = accept(s, (struct sockaddr*)&addr, &addr_len);
if(newfd == -1) {
#ifndef USE_WINSOCK
if(errno == EAGAIN || errno == EINTR)
return;
fatal_exit("accept: %s", strerror(errno));
#else
if(WSAGetLastError() == WSAEWOULDBLOCK ||
WSAGetLastError() == WSAEINPROGRESS ||
WSAGetLastError() == WSAECONNRESET)
return;
fatal_exit("accept: %s", wsa_strerror(WSAGetLastError()));
#endif
}
p = (struct tcp_proxy*)calloc(1, sizeof(*p));
if(!p) fatal_exit("out of memory");
memmove(&p->addr, &addr, addr_len);
p->addr_len = addr_len;
log_addr(1, "new tcp proxy", &p->addr, p->addr_len);
p->client_s = newfd;
p->server_s = socket(addr_is_ip6(srv_addr, srv_len)?AF_INET6:AF_INET,
SOCK_STREAM, 0);
if(p->server_s == -1) {
#ifndef USE_WINSOCK
fatal_exit("tcp socket: %s", strerror(errno));
#else
fatal_exit("tcp socket: %s", wsa_strerror(WSAGetLastError()));
#endif
}
fd_set_nonblock(p->client_s);
fd_set_nonblock(p->server_s);
if(connect(p->server_s, (struct sockaddr*)srv_addr, srv_len) == -1) {
#ifndef USE_WINSOCK
if(errno != EINPROGRESS) {
log_err("tcp connect: %s", strerror(errno));
close(p->server_s);
close(p->client_s);
#else
if(WSAGetLastError() != WSAEWOULDBLOCK &&
WSAGetLastError() != WSAEINPROGRESS) {
log_err("tcp connect: %s",
wsa_strerror(WSAGetLastError()));
closesocket(p->server_s);
closesocket(p->client_s);
#endif
free(p);
return;
}
}
p->timeout = *now;
dl_tv_add(&p->timeout, tcp_timeout);
/* listen to client and server */
FD_SET(FD_SET_T p->client_s, rorig);
FD_SET(FD_SET_T p->server_s, rorig);
if(p->client_s+1 > *max)
*max = p->client_s+1;
if(p->server_s+1 > *max)
*max = p->server_s+1;
/* add into proxy list */
p->next = *proxies;
*proxies = p;
}
/** relay TCP, read a part */
static int
tcp_relay_read(int s, struct tcp_send_list** first,
struct tcp_send_list** last, struct timeval* now,
struct timeval* delay, sldns_buffer* pkt)
{
struct tcp_send_list* item;
ssize_t r = recv(s, (void*)sldns_buffer_begin(pkt),
sldns_buffer_capacity(pkt), 0);
if(r == -1) {
#ifndef USE_WINSOCK
if(errno == EINTR || errno == EAGAIN)
return 1;
log_err("tcp read: %s", strerror(errno));
#else
if(WSAGetLastError() == WSAEINPROGRESS ||
WSAGetLastError() == WSAEWOULDBLOCK)
return 1;
log_err("tcp read: %s", wsa_strerror(WSAGetLastError()));
#endif
return 0;
} else if(r == 0) {
/* connection closed */
return 0;
}
item = (struct tcp_send_list*)malloc(sizeof(*item));
if(!item) {
log_err("out of memory");
return 0;
}
verbose(1, "read item len %d", (int)r);
item->len = (size_t)r;
item->item = memdup(sldns_buffer_begin(pkt), item->len);
if(!item->item) {
free(item);
log_err("out of memory");
return 0;
}
item->done = 0;
item->wait = *now;
dl_tv_add(&item->wait, delay);
item->next = NULL;
/* link in */
if(*first) {
(*last)->next = item;
} else {
*first = item;
}
*last = item;
return 1;
}
/** relay TCP, write a part */
static int
tcp_relay_write(int s, struct tcp_send_list** first,
struct tcp_send_list** last, struct timeval* now)
{
ssize_t r;
struct tcp_send_list* p;
while(*first) {
p = *first;
/* is the item ready? */
if(!dl_tv_smaller(&p->wait, now))
return 1;
/* write it */
r = send(s, (void*)(p->item + p->done), p->len - p->done, 0);
if(r == -1) {
#ifndef USE_WINSOCK
if(errno == EAGAIN || errno == EINTR)
return 1;
log_err("tcp write: %s", strerror(errno));
#else
if(WSAGetLastError() == WSAEWOULDBLOCK ||
WSAGetLastError() == WSAEINPROGRESS)
return 1;
log_err("tcp write: %s",
wsa_strerror(WSAGetLastError()));
#endif
return 0;
} else if(r == 0) {
/* closed */
return 0;
}
/* account it */
p->done += (size_t)r;
verbose(1, "write item %d of %d", (int)p->done, (int)p->len);
if(p->done >= p->len) {
free(p->item);
*first = p->next;
if(!*first)
*last = NULL;
free(p);
} else {
/* partial write */
return 1;
}
}
return 1;
}
int
create_tcp_accept_sock(struct addrinfo *addr, int v6only, int* noproto)
{
int s;
#if defined(SO_REUSEADDR) || defined(IPV6_V6ONLY)
int on = 1;
#endif /* SO_REUSEADDR || IPV6_V6ONLY */
verbose_print_addr(addr);
*noproto = 0;
if((s = socket(addr->ai_family, addr->ai_socktype, 0)) == -1) {
#ifndef USE_WINSOCK
if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
*noproto = 1;
return -1;
}
log_err("can't create socket: %s", strerror(errno));
#else
if(WSAGetLastError() == WSAEAFNOSUPPORT ||
WSAGetLastError() == WSAEPROTONOSUPPORT) {
*noproto = 1;
return -1;
}
log_err("can't create socket: %s",
wsa_strerror(WSAGetLastError()));
#endif
return -1;
}
#ifdef SO_REUSEADDR
if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
(socklen_t)sizeof(on)) < 0) {
#ifndef USE_WINSOCK
log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
strerror(errno));
close(s);
#else
log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
#endif
return -1;
}
#endif /* SO_REUSEADDR */
#if defined(IPV6_V6ONLY)
if(addr->ai_family == AF_INET6 && v6only) {
if(setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
(void*)&on, (socklen_t)sizeof(on)) < 0) {
#ifndef USE_WINSOCK
log_err("setsockopt(..., IPV6_V6ONLY, ...) failed: %s",
strerror(errno));
close(s);
#else
log_err("setsockopt(..., IPV6_V6ONLY, ...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
#endif
return -1;
}
}
#else
(void)v6only;
#endif /* IPV6_V6ONLY */
if(bind(s, addr->ai_addr, addr->ai_addrlen) != 0) {
#ifndef USE_WINSOCK
/* detect freebsd jail with no ipv6 permission */
if(addr->ai_family==AF_INET6 && errno==EINVAL)
*noproto = 1;
else {
log_err("can't bind socket: %s", strerror(errno));
log_addr(0, "failed address",
(struct sockaddr_storage*)addr->ai_addr,
addr->ai_addrlen);
}
close(s);
#else
log_err("can't bind socket: %s",
wsa_strerror(WSAGetLastError()));
log_addr(0, "failed address",
(struct sockaddr_storage*)addr->ai_addr,
addr->ai_addrlen);
closesocket(s);
#endif
return -1;
}
if(!fd_set_nonblock(s)) {
#ifndef USE_WINSOCK
close(s);
#else
closesocket(s);
#endif
return -1;
}
if(listen(s, TCP_BACKLOG) == -1) {
#ifndef USE_WINSOCK
log_err("can't listen: %s", strerror(errno));
close(s);
#else
log_err("can't listen: %s", wsa_strerror(WSAGetLastError()));
closesocket(s);
#endif
return -1;
}
return s;
}
int
create_udp_sock(int family, int socktype, struct sockaddr* addr,
socklen_t addrlen, int v6only, int* inuse, int* noproto,
int rcv, int snd, int listen, int* reuseport)
{
int s;
#if defined(SO_REUSEADDR) || defined(SO_REUSEPORT) || defined(IPV6_USE_MIN_MTU)
int on=1;
#endif
#ifdef IPV6_MTU
int mtu = IPV6_MIN_MTU;
#endif
#if !defined(SO_RCVBUFFORCE) && !defined(SO_RCVBUF)
(void)rcv;
#endif
#if !defined(SO_SNDBUFFORCE) && !defined(SO_SNDBUF)
(void)snd;
#endif
#ifndef IPV6_V6ONLY
(void)v6only;
#endif
if((s = socket(family, socktype, 0)) == -1) {
*inuse = 0;
#ifndef USE_WINSOCK
if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
*noproto = 1;
return -1;
}
log_err("can't create socket: %s", strerror(errno));
#else
if(WSAGetLastError() == WSAEAFNOSUPPORT ||
WSAGetLastError() == WSAEPROTONOSUPPORT) {
*noproto = 1;
return -1;
}
log_err("can't create socket: %s",
wsa_strerror(WSAGetLastError()));
#endif
*noproto = 0;
return -1;
}
if(listen) {
#ifdef SO_REUSEADDR
if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
(socklen_t)sizeof(on)) < 0) {
#ifndef USE_WINSOCK
log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
strerror(errno));
if(errno != ENOSYS) {
close(s);
*noproto = 0;
*inuse = 0;
return -1;
}
#else
log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
*noproto = 0;
*inuse = 0;
return -1;
#endif
}
#endif /* SO_REUSEADDR */
#ifdef SO_REUSEPORT
/* try to set SO_REUSEPORT so that incoming
* queries are distributed evenly among the receiving threads.
* Each thread must have its own socket bound to the same port,
* with SO_REUSEPORT set on each socket.
*/
if (reuseport && *reuseport &&
setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void*)&on,
(socklen_t)sizeof(on)) < 0) {
#ifdef ENOPROTOOPT
if(errno != ENOPROTOOPT || verbosity >= 3)
log_warn("setsockopt(.. SO_REUSEPORT ..) failed: %s",
strerror(errno));
#endif
/* this option is not essential, we can continue */
*reuseport = 0;
}
#else
(void)reuseport;
#endif /* defined(SO_REUSEPORT) */
}
if(rcv) {
#ifdef SO_RCVBUF
int got;
socklen_t slen = (socklen_t)sizeof(got);
# ifdef SO_RCVBUFFORCE
/* Linux specific: try to use root permission to override
* system limits on rcvbuf. The limit is stored in
* /proc/sys/net/core/rmem_max or sysctl net.core.rmem_max */
if(setsockopt(s, SOL_SOCKET, SO_RCVBUFFORCE, (void*)&rcv,
(socklen_t)sizeof(rcv)) < 0) {
if(errno != EPERM) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., SO_RCVBUFFORCE, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., SO_RCVBUFFORCE, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
# endif /* SO_RCVBUFFORCE */
if(setsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&rcv,
(socklen_t)sizeof(rcv)) < 0) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., SO_RCVBUF, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., SO_RCVBUF, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
/* check if we got the right thing or if system
* reduced to some system max. Warn if so */
if(getsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&got,
&slen) >= 0 && got < rcv/2) {
log_warn("so-rcvbuf %u was not granted. "
"Got %u. To fix: start with "
"root permissions(linux) or sysctl "
"bigger net.core.rmem_max(linux) or "
"kern.ipc.maxsockbuf(bsd) values.",
(unsigned)rcv, (unsigned)got);
}
# ifdef SO_RCVBUFFORCE
}
# endif
#endif /* SO_RCVBUF */
}
/* first do RCVBUF as the receive buffer is more important */
if(snd) {
#ifdef SO_SNDBUF
int got;
socklen_t slen = (socklen_t)sizeof(got);
# ifdef SO_SNDBUFFORCE
/* Linux specific: try to use root permission to override
* system limits on sndbuf. The limit is stored in
* /proc/sys/net/core/wmem_max or sysctl net.core.wmem_max */
if(setsockopt(s, SOL_SOCKET, SO_SNDBUFFORCE, (void*)&snd,
(socklen_t)sizeof(snd)) < 0) {
if(errno != EPERM) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., SO_SNDBUFFORCE, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., SO_SNDBUFFORCE, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
# endif /* SO_SNDBUFFORCE */
if(setsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&snd,
(socklen_t)sizeof(snd)) < 0) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., SO_SNDBUF, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., SO_SNDBUF, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
/* check if we got the right thing or if system
* reduced to some system max. Warn if so */
if(getsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&got,
&slen) >= 0 && got < snd/2) {
log_warn("so-sndbuf %u was not granted. "
"Got %u. To fix: start with "
"root permissions(linux) or sysctl "
"bigger net.core.wmem_max(linux) or "
"kern.ipc.maxsockbuf(bsd) values.",
(unsigned)snd, (unsigned)got);
}
# ifdef SO_SNDBUFFORCE
}
# endif
#endif /* SO_SNDBUF */
}
if(family == AF_INET6) {
# if defined(IPV6_V6ONLY)
if(v6only) {
int val=(v6only==2)?0:1;
if (setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
(void*)&val, (socklen_t)sizeof(val)) < 0) {
#ifndef USE_WINSOCK
log_err("setsockopt(..., IPV6_V6ONLY"
", ...) failed: %s", strerror(errno));
close(s);
#else
log_err("setsockopt(..., IPV6_V6ONLY"
", ...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
#endif
*noproto = 0;
*inuse = 0;
return -1;
}
}
# endif
# if defined(IPV6_USE_MIN_MTU)
/*
* There is no fragmentation of IPv6 datagrams
* during forwarding in the network. Therefore
* we do not send UDP datagrams larger than
* the minimum IPv6 MTU of 1280 octets. The
* EDNS0 message length can be larger if the
* network stack supports IPV6_USE_MIN_MTU.
*/
if (setsockopt(s, IPPROTO_IPV6, IPV6_USE_MIN_MTU,
(void*)&on, (socklen_t)sizeof(on)) < 0) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., IPV6_USE_MIN_MTU, "
"...) failed: %s", strerror(errno));
close(s);
# else
log_err("setsockopt(..., IPV6_USE_MIN_MTU, "
"...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
# elif defined(IPV6_MTU)
/*
* On Linux, to send no larger than 1280, the PMTUD is
* disabled by default for datagrams anyway, so we set
* the MTU to use.
*/
if (setsockopt(s, IPPROTO_IPV6, IPV6_MTU,
(void*)&mtu, (socklen_t)sizeof(mtu)) < 0) {
# ifndef USE_WINSOCK
log_err("setsockopt(..., IPV6_MTU, ...) failed: %s",
strerror(errno));
close(s);
# else
log_err("setsockopt(..., IPV6_MTU, ...) failed: %s",
wsa_strerror(WSAGetLastError()));
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
# endif /* IPv6 MTU */
} else if(family == AF_INET) {
# if defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
/* linux 3.15 has IP_PMTUDISC_OMIT, Hannes Frederic Sowa made it so that
* PMTU information is not accepted, but fragmentation is allowed
* if and only if the packet size exceeds the outgoing interface MTU
* (and also uses the interface mtu to determine the size of the packets).
* So there won't be any EMSGSIZE error. Against DNS fragmentation attacks.
* FreeBSD already has same semantics without setting the option. */
int omit_set = 0;
int action;
# if defined(IP_PMTUDISC_OMIT)
action = IP_PMTUDISC_OMIT;
if (setsockopt(s, IPPROTO_IP, IP_MTU_DISCOVER,
&action, (socklen_t)sizeof(action)) < 0) {
if (errno != EINVAL) {
log_err("setsockopt(..., IP_MTU_DISCOVER, IP_PMTUDISC_OMIT...) failed: %s",
strerror(errno));
# ifndef USE_WINSOCK
close(s);
# else
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
}
else
{
omit_set = 1;
}
# endif
if (omit_set == 0) {
action = IP_PMTUDISC_DONT;
if (setsockopt(s, IPPROTO_IP, IP_MTU_DISCOVER,
&action, (socklen_t)sizeof(action)) < 0) {
log_err("setsockopt(..., IP_MTU_DISCOVER, IP_PMTUDISC_DONT...) failed: %s",
strerror(errno));
# ifndef USE_WINSOCK
close(s);
# else
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
}
# elif defined(IP_DONTFRAG)
int off = 0;
if (setsockopt(s, IPPROTO_IP, IP_DONTFRAG,
&off, (socklen_t)sizeof(off)) < 0) {
log_err("setsockopt(..., IP_DONTFRAG, ...) failed: %s",
strerror(errno));
# ifndef USE_WINSOCK
close(s);
# else
closesocket(s);
# endif
*noproto = 0;
*inuse = 0;
return -1;
}
# endif /* IPv4 MTU */
}
if(bind(s, (struct sockaddr*)addr, addrlen) != 0) {
*noproto = 0;
*inuse = 0;
#ifndef USE_WINSOCK
#ifdef EADDRINUSE
*inuse = (errno == EADDRINUSE);
/* detect freebsd jail with no ipv6 permission */
if(family==AF_INET6 && errno==EINVAL)
*noproto = 1;
else if(errno != EADDRINUSE) {
log_err_addr("can't bind socket", strerror(errno),
(struct sockaddr_storage*)addr, addrlen);
}
#endif /* EADDRINUSE */
close(s);
#else /* USE_WINSOCK */
if(WSAGetLastError() != WSAEADDRINUSE &&
WSAGetLastError() != WSAEADDRNOTAVAIL) {
log_err_addr("can't bind socket",
wsa_strerror(WSAGetLastError()),
(struct sockaddr_storage*)addr, addrlen);
}
closesocket(s);
#endif
return -1;
}
if(!fd_set_nonblock(s)) {
*noproto = 0;
*inuse = 0;
#ifndef USE_WINSOCK
close(s);
#else
closesocket(s);
#endif
return -1;
}
return s;
}