ssize_t recvfrom(int sockfd, void *buf, size_t len, int flags,
struct sockaddr *src_addr, socklen_t *addrlen)
{
struct fdtab_entry *e = fdtab_get(sockfd);
switch(e->type) {
case FDTAB_TYPE_UNIX_SOCKET:
assert(!"NYI");
return -1;
break;
case FDTAB_TYPE_LWIP_SOCKET:
lwip_mutex_lock();
ssize_t ret = lwip_recvfrom(e->fd, buf, len, flags, src_addr, addrlen);
lwip_mutex_unlock();
return ret;
case FDTAB_TYPE_AVAILABLE:
errno = EBADF;
return -1;
default:
errno = ENOTSOCK;
return -1;
}
}
int listen(int sockfd, int backlog)
{
struct fdtab_entry *e = fdtab_get(sockfd);
switch(e->type) {
case FDTAB_TYPE_UNIX_SOCKET:
POSIXCOMPAT_DEBUG("listen(%d, %d)\n", sockfd, backlog);
struct _unix_socket *us = e->handle;
errval_t err =
unixsock_export(us, unixsock_listening, unixsock_connected,
get_default_waitset(), IDC_EXPORT_FLAGS_DEFAULT);
if(err_is_fail(err)) {
DEBUG_ERR(err, "unixsock_export failed");
return -1;
}
while(us->u.passive.listen_iref == NULL_IREF) {
// XXX: Should wait only on monitor
event_dispatch(get_default_waitset());
}
us->passive = true;
us->u.passive.max_backlog = backlog;
us->u.passive.backlog = calloc(backlog, sizeof(struct unixsock_binding *));
char str[128];
snprintf(str, 128, "%"PRIuIREF, us->u.passive.listen_iref);
err = vfs_write(us->vfs_handle, str, strlen(str), NULL);
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "vfs_write");
}
break;
case FDTAB_TYPE_LWIP_SOCKET:
lwip_mutex_lock();
int ret = lwip_listen(e->fd, backlog);
lwip_mutex_unlock();
return ret;
default:
return -1;
}
return 0;
}
int bind(int sockfd, const struct sockaddr *addr, socklen_t addrlen)
{
struct fdtab_entry *e = fdtab_get(sockfd);
switch(e->type) {
case FDTAB_TYPE_UNIX_SOCKET:
assert(addrlen >= sizeof(struct sockaddr_un));
assert(addr->sa_family == AF_UNIX);
struct _unix_socket *us = e->handle;
memcpy(&us->sockaddr, addr, sizeof(struct sockaddr_un));
POSIXCOMPAT_DEBUG("bind(%d, %p (%s), %u)\n", sockfd, addr,
us->sockaddr.sun_path, addrlen);
// XXX: Should fail if file already exists
// Create socket file
errval_t err = vfs_create(us->sockaddr.sun_path, &us->vfs_handle);
if(err_is_fail(err)) {
DEBUG_ERR(err, "vfs_create");
return -1;
}
break;
case FDTAB_TYPE_LWIP_SOCKET:
lwip_mutex_lock();
int ret = lwip_bind(e->fd, addr, addrlen);
lwip_mutex_unlock();
return ret;
default:
return -1;
}
return 0;
}
int accept(int sockfd, struct sockaddr *addr, socklen_t *addrlen)
{
struct fdtab_entry *e = fdtab_get(sockfd);
POSIXCOMPAT_DEBUG("%s:%s:%d: accept(sockfd=%d , e->type=%d)\n",
__FILE__, __FUNCTION__, __LINE__, sockfd, e->type);
switch(e->type) {
case FDTAB_TYPE_UNIX_SOCKET:
{
struct _unix_socket *us = e->handle;
int i;
if(!us->passive) {
return -1;
}
for(;;) {
for(i = 0; i < us->u.passive.max_backlog; i++) {
if(us->u.passive.backlog[i] != NULL) {
break;
}
}
if(i != us->u.passive.max_backlog) {
// Found a pending connection
break;
}
if(us->nonblocking) {
errno = EAGAIN;
return -1;
}
// XXX: This should wait only on the monitor
event_dispatch(get_default_waitset());
}
// XXX: Using socket() to create new socket. Probably dangerous.
int newfd = socket(AF_UNIX, us->type, us->protocol);
assert(newfd != -1);
struct _unix_socket *newus = fdtab_get(newfd)->handle;
newus->u.active.binding = us->u.passive.backlog[i];
newus->u.active.mode = _UNIX_SOCKET_MODE_CONNECTED;
memcpy(&newus->sockaddr, &us->sockaddr, sizeof(struct sockaddr_un));
// Associate binding with new socket and remove from backlog
assert(newus->u.active.binding->st == NULL);
newus->u.active.binding->st = newus;
us->u.passive.backlog[i] = NULL;
if(addr != NULL) {
assert(addrlen != NULL);
// TODO: Should request address from peer (if peer is bound)
if(*addrlen > sizeof(struct sockaddr_un)) {
*addrlen = sizeof(struct sockaddr_un);
}
memcpy(addr, &newus->peer, *addrlen);
}
return newfd;
}
case FDTAB_TYPE_LWIP_SOCKET:
{
lwip_mutex_lock();
int newfd = lwip_accept(e->fd, addr, addrlen);
lwip_mutex_unlock();
if(newfd != -1) {
struct fdtab_entry newe;
newe.type = FDTAB_TYPE_LWIP_SOCKET;
newe.fd = newfd;
newe.inherited = false;
newe.epoll_fd = -1;
newfd = fdtab_alloc(&newe);
POSIXCOMPAT_DEBUG("accept(%d, _, _) as fd %d\n", sockfd, newfd);
if (newfd < 0) {
return -1;
}
}
return newfd;
}
default:
return -1;
}
}
int socket(int domain, int type, int protocol)
{
struct fdtab_entry e;
switch(domain) {
case AF_UNIX:
if(type != SOCK_STREAM) {
errno = EPROTOTYPE;
return -1;
}
if(protocol != 0) {
errno = EPROTONOSUPPORT;
return -1;
}
struct _unix_socket *us = malloc(sizeof(struct _unix_socket));
assert(us != NULL);
memset(us, 0, sizeof(struct _unix_socket));
us->type = type;
us->protocol = protocol;
us->recv_buf_size = _UNIX_SOCKET_RECV_BUF_SIZE;
/*
* XXX: Even though POSIX says the result of
* get{sock,peer}name() shall be unspecified if not bound,
* some programs still expect its type to be of AF_UNIX.
*/
us->sockaddr.sun_len = us->peer.sun_len = sizeof(struct sockaddr_un);
us->sockaddr.sun_family = us->peer.sun_family = AF_UNIX;
e.type = FDTAB_TYPE_UNIX_SOCKET;
e.handle = us;
e.inherited = false;
e.epoll_fd = -1;
break;
case AF_INET:
{
lwip_mutex_lock();
int fd = lwip_socket(domain, type, protocol);
lwip_mutex_unlock();
if(fd == -1) {
return fd;
}
e.type = FDTAB_TYPE_LWIP_SOCKET;
e.fd = fd;
e.inherited = false;
e.epoll_fd = -1;
}
break;
default:
errno = EAFNOSUPPORT;
return -1;
}
int fd = fdtab_alloc(&e);
POSIXCOMPAT_DEBUG("socket(%d, %d, %d) as fd %d\n", domain, type, protocol, fd);
if (fd < 0) {
return -1;
} else {
return fd;
}
}
ssize_t recv(int sockfd, void *buf, size_t len, int flags)
{
struct fdtab_entry *e = fdtab_get(sockfd);
switch(e->type) {
case FDTAB_TYPE_UNIX_SOCKET:
{
struct _unix_socket *us = e->handle;
// XXX: Don't support flags
assert(flags == 0);
thread_mutex_lock(&us->mutex);
if(us->passive
|| us->u.active.mode != _UNIX_SOCKET_MODE_CONNECTED) {
errno = ENOTCONN;
thread_mutex_unlock(&us->mutex);
return -1;
}
if(us->recv_buf_valid == 0) {
// No more data
if(us->nonblocking) {
errno = EAGAIN;
thread_mutex_unlock(&us->mutex);
return -1;
} else {
struct waitset ws;
errval_t err;
waitset_init(&ws);
err = us->u.active.binding->change_waitset
(us->u.active.binding, &ws);
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "change_waitset");
}
while(us->recv_buf_valid == 0) {
err = event_dispatch(&ws);
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "waitset_destroy");
}
}
// XXX: Assume it was on the default waitset
err = us->u.active.binding->change_waitset
(us->u.active.binding, get_default_waitset());
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "change_waitset");
}
err = waitset_destroy(&ws);
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "waitset_destroy");
}
}
}
size_t recved = 0;
while(recved < len && us->recv_list != NULL) {
struct _unix_socket_recv *usr = us->recv_list;
size_t consume = MIN(len - recved, usr->size - usr->consumed);
memcpy(buf + recved, &usr->msg[usr->consumed], consume);
usr->consumed += consume;
us->recv_buf_valid -= consume;
recved += consume;
if(usr->consumed == usr->size) {
us->recv_list = usr->next;
if(us->recv_list == NULL) {
us->recv_list_end = NULL;
}
free(usr->msg);
free(usr);
} else {
assert(recved == len);
}
}
thread_mutex_unlock(&us->mutex);
return recved;
}
case FDTAB_TYPE_LWIP_SOCKET:
lwip_mutex_lock();
ssize_t ret = lwip_recv(e->fd, buf, len, flags);
lwip_mutex_unlock();
return ret;
case FDTAB_TYPE_AVAILABLE:
errno = EBADF;
return -1;
default:
errno = ENOTSOCK;
return -1;
}
}
ssize_t send(int sockfd, const void *buf, size_t len, int flags)
{
struct fdtab_entry *e = fdtab_get(sockfd);
switch(e->type) {
case FDTAB_TYPE_UNIX_SOCKET:
{
struct _unix_socket *us = e->handle;
// XXX: Don't support flags
assert(flags == 0);
thread_mutex_lock(&us->mutex);
if(us->passive
|| us->u.active.mode != _UNIX_SOCKET_MODE_CONNECTED) {
errno = ENOTCONN;
thread_mutex_unlock(&us->mutex);
return -1;
}
if(us->send_buf != NULL) {
if(us->nonblocking) {
errno = EAGAIN;
thread_mutex_unlock(&us->mutex);
return -1;
} else {
assert(!"NYI");
}
}
// Bleh. Gotta copy here. I can't just wait until the
// message is fully sent, as that might block
// indefinitely.
us->send_buf = malloc(len);
memcpy(us->send_buf, buf, len);
struct event_closure ec = {
.handler = unixsock_sent,
.arg = us,
};
errval_t err = us->u.active.binding->tx_vtbl.
send(us->u.active.binding, ec, us->send_buf, len);
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "unixsock->send");
thread_mutex_unlock(&us->mutex);
return -1;
}
// Wait until all data sent if blocking
if(!us->nonblocking) {
struct waitset ws;
waitset_init(&ws);
err = us->u.active.binding->change_waitset
(us->u.active.binding, &ws);
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "change_waitset");
}
while(us->send_buf != NULL) {
err = event_dispatch(&ws);
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "waitset_destroy");
}
}
// XXX: Assume it was on the default waitset
err = us->u.active.binding->change_waitset
(us->u.active.binding, get_default_waitset());
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "change_waitset");
}
err = waitset_destroy(&ws);
if(err_is_fail(err)) {
USER_PANIC_ERR(err, "waitset_destroy");
}
}
// XXX: We send all or nothing
thread_mutex_unlock(&us->mutex);
return len;
}
case FDTAB_TYPE_LWIP_SOCKET:
lwip_mutex_lock();
ssize_t ret = lwip_send(e->fd, buf, len, flags);
lwip_mutex_unlock();
return ret;
case FDTAB_TYPE_AVAILABLE:
errno = EBADF;
return -1;
default:
errno = ENOTSOCK;
return -1;
}
}
int epoll_wait(int epfd, struct epoll_event *events,
int maxevents, int timeout)
{
struct fdtab_entry *mye = fdtab_get(epfd);
assert(mye->type == FDTAB_TYPE_EPOLL_INSTANCE);
struct _epoll_fd *efd = mye->handle;
struct monitor_binding *mb = get_monitor_binding();
errval_t err;
/* waitset_init(&efd->ws); */
assert(maxevents >= 1);
for(struct _epoll_events_list *i = efd->events; i != NULL; i = i->next) {
struct fdtab_entry *e = fdtab_get(i->fd);
struct epoll_event *event = &i->event;
switch (e->type) {
case FDTAB_TYPE_LWIP_SOCKET:
{
int retval;
lwip_mutex_lock();
if(event->events & EPOLLIN) {
retval = lwip_sock_waitset_register_read(e->fd, &efd->ws);
assert(retval == 0);
}
if(event->events & EPOLLOUT) {
retval = lwip_sock_waitset_register_write(e->fd, &efd->ws);
assert(retval == 0);
}
lwip_mutex_unlock();
}
break;
case FDTAB_TYPE_UNIX_SOCKET:
{
struct _unix_socket *us = e->handle;
if(event->events & EPOLLIN) {
if (us->passive) { /* passive side */
int j;
/* Check for pending connection requests. */
for (j = 0; j < us->u.passive.max_backlog; j++)
{
if (us->u.passive.backlog[j] != NULL) {
break;
}
}
/*
* If there are not pending connection request
* wait on monitor binding.
*/
if (j == us->u.passive.max_backlog) {
/* wait on monitor */
err = mb->change_waitset(mb, &efd->ws);
if (err_is_fail(err)) {
USER_PANIC_ERR(err, "change_waitset");
}
}
}
}
if(event->events & EPOLLOUT) {
assert(!us->passive);
if(us->u.active.mode == _UNIX_SOCKET_MODE_CONNECTING) {
/* wait on monitor */
err = mb->change_waitset(mb, &efd->ws);
if (err_is_fail(err)) {
USER_PANIC_ERR(err, "change_waitset");
}
}
}
assert(event->events & (EPOLLIN | EPOLLOUT));
// Change waitset
err = us->u.active.binding->change_waitset
(us->u.active.binding, &efd->ws);
if (err_is_fail(err)) {
USER_PANIC_ERR(err, "change waitset");
}
}
break;
default:
{
fprintf(stderr, "change waitset on FD type %d NYI.\n",
e->type);
assert(!"NYI");
errno = EBADF;
return -1;
}
}
}
// Timeout handling
struct timeout_event toe = {
.fired = false
};
struct deferred_event timeout_event;
if (timeout > 0) {
deferred_event_init(&timeout_event);
err = deferred_event_register(&timeout_event, &efd->ws, timeout,
MKCLOSURE(timeout_fired, &toe));
if (err_is_fail(err)) {
errno = EINVAL;
return -1;
}
}
int retevents = 0;
while(!toe.fired && retevents == 0) {
if(timeout == 0) {
// Just poll once, don't block
err = event_dispatch_non_block(&efd->ws);
assert(err_is_ok(err) || err_no(err) == LIB_ERR_NO_EVENT);
toe.fired = true;
} else {
err = event_dispatch(&efd->ws);
if (err_is_fail(err)) {
USER_PANIC_ERR(err, "Error in event_dispatch.");
}
}
// Return ready file descriptors
for(struct _epoll_events_list *i = efd->events; i != NULL; i = i->next) {
struct epoll_event *event = &i->event;
struct fdtab_entry *e = fdtab_get(i->fd);
assert(retevents < maxevents);
events[retevents] = *event;
events[retevents].events = 0;
// Check errors (hangup)
{
switch (e->type) {
case FDTAB_TYPE_LWIP_SOCKET:
{
lwip_mutex_lock();
if (!lwip_sock_is_open(e->fd)) {
events[retevents].events |= EPOLLHUP;
}
lwip_mutex_unlock();
}
break;
default:
// No-Op
break;
}
}
// Check readable FDs
if(event->events & EPOLLIN) {
switch (e->type) {
case FDTAB_TYPE_LWIP_SOCKET:
{
lwip_mutex_lock();
if (lwip_sock_ready_read(e->fd)) {
events[retevents].events |= EPOLLIN;
}
lwip_mutex_unlock();
}
break;
case FDTAB_TYPE_UNIX_SOCKET:
{
struct _unix_socket *us = e->handle;
if (us->passive) { /* passive side */
/* Check for pending connection requests. */
for (int j = 0; j < us->u.passive.max_backlog; j++)
{
if (us->u.passive.backlog[j] != NULL) {
events[retevents].events |= EPOLLIN;
break;
}
}
} else { /* active side */
/* Check for incoming data. */
if (us->recv_buf_valid > 0) {
events[retevents].events |= EPOLLIN;
}
}
}
break;
default:
{
fprintf(stderr, "epoll_wait() on FD type %d NYI.\n",
e->type);
assert(!"NYI");
errno = EBADF;
return -1;
}
}
}
// Check writeable FDs
if(event->events & EPOLLOUT) {
switch (e->type) {
case FDTAB_TYPE_LWIP_SOCKET:
{
lwip_mutex_lock();
if (lwip_sock_ready_write(e->fd)) {
events[retevents].events |= EPOLLOUT;
}
lwip_mutex_unlock();
}
break;
case FDTAB_TYPE_UNIX_SOCKET:
{
struct _unix_socket *us = e->handle;
assert(!us->passive);
switch (us->u.active.mode) {
case _UNIX_SOCKET_MODE_CONNECTING:
break;
case _UNIX_SOCKET_MODE_CONNECTED:
if (us->send_buf == NULL) {
events[retevents].events |= EPOLLOUT;
}
break;
}
}
break;
default:
{
fprintf(stderr, "epoll_wait() on FD type %d NYI.\n",
e->type);
assert(!"NYI");
errno = EBADF;
return -1;
}
}
}
// If any events were returned, go to next entry in array
if(events[retevents].events != 0) {
retevents++;
}
}
}
// Remove timeout from waitset if it was set and not fired
if(timeout > 0 && !toe.fired) {
deferred_event_cancel(&timeout_event);
}
// Restore old waitsets
for(struct _epoll_events_list *i = efd->events; i != NULL; i = i->next) {
struct fdtab_entry *e = fdtab_get(i->fd);
struct epoll_event *event = &i->event;
switch (e->type) {
case FDTAB_TYPE_LWIP_SOCKET:
{
lwip_mutex_lock();
if(event->events & EPOLLIN) {
err = lwip_sock_waitset_deregister_read(e->fd);
if (err_is_fail(err) &&
err_no(err) != LIB_ERR_CHAN_NOT_REGISTERED) {
USER_PANIC_ERR(err, "error deregister read channel for "
"lwip socket");
}
}
if(event->events & EPOLLOUT) {
err = lwip_sock_waitset_deregister_write(e->fd);
if (err_is_fail(err) &&
err_no(err) != LIB_ERR_CHAN_NOT_REGISTERED) {
USER_PANIC_ERR(err, "error deregister write channel for "
"lwip socket");
}
}
lwip_mutex_unlock();
}
break;
case FDTAB_TYPE_UNIX_SOCKET:
{
// NYI
}
break;
default:
{
fprintf(stderr, "change waitset on FD type %d NYI.\n",
e->type);
assert(!"NYI");
errno = EBADF;
return -1;
}
}
}
return retevents;
}
int epoll_pwait(int epfd, struct epoll_event *events,
int maxevents, int timeout,
const sigset_t *sigmask)
{
assert(!"NYI");
}
static void call_tcp_tmr(void)
{
lwip_mutex_lock();
tcp_tmr();
lwip_mutex_unlock();
}
