int
close(int fd)
{
int ret;
/* This is a cancelation point: */
_thread_enter_cancellation_point();
if ((fd < 0) || (fd >= _thread_max_fdtsize) ||
(fd == _thread_kern_pipe[0]) || (fd == _thread_kern_pipe[1])) {
errno = EBADF;
ret = -1;
} else if ((ret = _FD_LOCK(fd, FD_RDWR_CLOSE, NULL)) != -1) {
/*
* We need to hold the entry spinlock till after
* _thread_sys_close() to stop races caused by the
* fd state transition.
*/
_SPINLOCK(&_thread_fd_table[fd]->lock);
_thread_fd_entry_close(fd);
/* Close the file descriptor: */
ret = _thread_sys_close(fd);
_SPINUNLOCK(&_thread_fd_table[fd]->lock);
_FD_UNLOCK(fd, FD_RDWR_CLOSE);
}
/* No longer in a cancellation point: */
_thread_leave_cancellation_point();
return (ret);
}
int
_sem_wait(sem_t *sem)
{
int retval;
_thread_enter_cancellation_point();
_SEM_CHECK_VALIDITY(sem);
pthread_mutex_lock(&(*sem)->lock);
while ((*sem)->count == 0) {
(*sem)->nwaiters++;
pthread_cond_wait(&(*sem)->gtzero, &(*sem)->lock);
(*sem)->nwaiters--;
}
(*sem)->count--;
pthread_mutex_unlock(&(*sem)->lock);
retval = 0;
RETURN:
_thread_leave_cancellation_point();
return retval;
}
unsigned int
sleep(unsigned int seconds)
{
unsigned int ret;
_thread_enter_cancellation_point();
ret = __sleep(seconds);
_thread_leave_cancellation_point();
return ret;
}
int
_pause(void)
{
int ret;
_thread_enter_cancellation_point();
ret = __pause();
_thread_leave_cancellation_point();
return ret;
}
int
__sigsuspend(const sigset_t * set)
{
int ret;
_thread_enter_cancellation_point();
ret = _sigsuspend(set);
_thread_leave_cancellation_point();
return ret;
}
ssize_t
writev(int fd, const struct iovec *iov, int iovcnt)
{
ssize_t ret;
_thread_enter_cancellation_point();
ret = _writev(fd, iov, iovcnt);
_thread_leave_cancellation_point();
return ret;
}
int
tcdrain(int fd)
{
int ret;
_thread_enter_cancellation_point();
ret = __tcdrain(fd);
_thread_leave_cancellation_point();
return ret;
}
int
poll(struct pollfd *fds, unsigned int nfds, int timeout)
{
int ret;
_thread_enter_cancellation_point();
ret = _poll(fds, nfds, timeout);
_thread_leave_cancellation_point();
return ret;
}
pid_t
_waitpid(pid_t wpid, int *status, int options)
{
pid_t ret;
_thread_enter_cancellation_point();
ret = __waitpid(wpid, status, options);
_thread_leave_cancellation_point();
return ret;
}
ssize_t
__sendmsg(int fd, const struct msghdr *msg, int flags)
{
int ret;
_thread_enter_cancellation_point();
ret = _sendmsg(fd, msg, flags);
_thread_leave_cancellation_point();
return (ret);
}
ssize_t
recvfrom(int fd, void *buf, size_t len, int flags, struct sockaddr * from, socklen_t *from_len)
{
struct pthread *curthread = _get_curthread();
ssize_t ret;
/* This is a cancellation point: */
_thread_enter_cancellation_point();
if ((ret = _FD_LOCK(fd, FD_READ, NULL)) == 0) {
while ((ret = _thread_sys_recvfrom(fd, buf, len, flags, from, from_len)) < 0) {
if (!(_thread_fd_table[fd]->status_flags->flags & O_NONBLOCK) &&
!(flags & MSG_DONTWAIT) &&
((errno == EWOULDBLOCK) || (errno == EAGAIN))) {
curthread->data.fd.fd = fd;
/* Set the timeout: */
_thread_kern_set_timeout(_FD_RCVTIMEO(fd));
curthread->interrupted = 0;
curthread->closing_fd = 0;
curthread->timeout = 0;
_thread_kern_sched_state(PS_FDR_WAIT, __FILE__, __LINE__);
/* Check if the wait was interrupted: */
if (curthread->interrupted) {
/* Return an error status: */
errno = EINTR;
ret = -1;
break;
} else if (curthread->closing_fd) {
/* Return an error status: */
errno = EBADF;
ret = -1;
break;
} else if (curthread->timeout) {
/* Return an error status: */
errno = EWOULDBLOCK;
ret = -1;
break;
}
} else {
ret = -1;
break;
}
}
_FD_UNLOCK(fd, FD_READ);
}
/* No longer in a cancellation point: */
_thread_leave_cancellation_point();
return (ret);
}
int
nanosleep(const struct timespec * time_to_sleep, struct timespec *
time_remaining)
{
int ret;
_thread_enter_cancellation_point();
ret = _nanosleep(time_to_sleep, time_remaining);
_thread_leave_cancellation_point();
return ret;
}
int
_aio_suspend(const struct aiocb * const iocbs[], int niocb, const struct
timespec *timeout)
{
int ret;
_thread_enter_cancellation_point();
ret = __sys_aio_suspend(iocbs, niocb, timeout);
_thread_leave_cancellation_point();
return ret;
}
ssize_t
preadv(int fd, const struct iovec * iov, int iovcnt, off_t offset)
{
ssize_t ret;
/* This is a cancellation point: */
_thread_enter_cancellation_point();
ret = _thread_sys_preadv(fd, iov, iovcnt, offset);
/* No longer in a cancellation point: */
_thread_leave_cancellation_point();
return (ret);
}
int
__msync(void *addr, size_t len, int flags)
{
int ret;
/*
* XXX This is quite pointless unless we know how to get the
* file descriptor associated with the memory, and lock it for
* write. The only real use of this wrapper is to guarantee
* a cancellation point, as per the standard. sigh.
*/
_thread_enter_cancellation_point();
ret = _msync(addr, len, flags);
_thread_leave_cancellation_point();
return ret;
}
/*
* Note: a thread calling wait4 may have its state changed to waiting
* until awakened by a signal. Also note that system(3), for example,
* blocks SIGCHLD and calls waitpid (which calls wait4). If the process
* started by system(3) doesn't finish before this function is called the
* function will never awaken -- system(3) also ignores SIGINT and SIGQUIT.
*
* Thus always unmask SIGCHLD here.
*/
pid_t
wait4(pid_t pid, int *istat, int options, struct rusage * rusage)
{
struct pthread *curthread = _get_curthread();
pid_t ret;
sigset_t mask, omask;
/* This is a cancellation point: */
_thread_enter_cancellation_point();
_thread_kern_sig_defer();
sigemptyset(&mask);
sigaddset(&mask, SIGCHLD);
sigprocmask(SIG_UNBLOCK, &mask, &omask);
/* Perform a non-blocking wait4 syscall: */
while ((ret = _thread_sys_wait4(pid, istat, options | WNOHANG, rusage)) == 0 && (options & WNOHANG) == 0) {
/* Reset the interrupted operation flag: */
curthread->interrupted = 0;
/* Schedule the next thread while this one waits: */
_thread_kern_sched_state(PS_WAIT_WAIT, __FILE__, __LINE__);
/* Check if this call was interrupted by a signal: */
if (curthread->interrupted) {
errno = EINTR;
ret = -1;
break;
}
}
sigprocmask(SIG_SETMASK, &omask, NULL);
_thread_kern_sig_undefer();
/* No longer in a cancellation point: */
_thread_leave_cancellation_point();
return (ret);
}
int
_pthread_cond_timedwait(pthread_cond_t * cond, pthread_mutex_t * mutex,
const struct timespec * abstime)
{
struct pthread *curthread = _get_curthread();
int rval = 0;
int done = 0;
int interrupted = 0;
int seqno;
_thread_enter_cancellation_point();
if (abstime == NULL || abstime->tv_sec < 0 || abstime->tv_nsec < 0 ||
abstime->tv_nsec >= 1000000000)
return (EINVAL);
/*
* If the condition variable is statically initialized, perform dynamic
* initialization.
*/
if (*cond == NULL && (rval = pthread_cond_init(cond, NULL)) != 0)
return (rval);
/*
* Enter a loop waiting for a condition signal or broadcast
* to wake up this thread. A loop is needed in case the waiting
* thread is interrupted by a signal to execute a signal handler.
* It is not (currently) possible to remain in the waiting queue
* while running a handler. Instead, the thread is interrupted
* and backed out of the waiting queue prior to executing the
* signal handler.
*/
do {
/* Lock the condition variable structure: */
_SPINLOCK(&(*cond)->lock);
/*
* If the condvar was statically allocated, properly
* initialize the tail queue.
*/
if (((*cond)->c_flags & COND_FLAGS_INITED) == 0) {
TAILQ_INIT(&(*cond)->c_queue);
(*cond)->c_flags |= COND_FLAGS_INITED;
}
/* Process according to condition variable type: */
switch ((*cond)->c_type) {
/* Fast condition variable: */
case COND_TYPE_FAST:
if ((mutex == NULL) || (((*cond)->c_mutex != NULL) &&
((*cond)->c_mutex != *mutex))) {
/* Return invalid argument error: */
rval = EINVAL;
/* Unlock the condition variable structure: */
_SPINUNLOCK(&(*cond)->lock);
} else {
/* Set the wakeup time: */
curthread->wakeup_time.tv_sec =
abstime->tv_sec;
curthread->wakeup_time.tv_nsec =
abstime->tv_nsec;
/* Reset the timeout and interrupted flags: */
curthread->timeout = 0;
curthread->interrupted = 0;
/*
* Queue the running thread for the condition
* variable:
*/
cond_queue_enq(*cond, curthread);
/* Remember the mutex and sequence number: */
(*cond)->c_mutex = *mutex;
seqno = (*cond)->c_seqno;
/* Unlock the mutex: */
if ((rval = _mutex_cv_unlock(mutex)) != 0) {
/*
* Cannot unlock the mutex, so remove
* the running thread from the condition
* variable queue:
*/
cond_queue_remove(*cond, curthread);
/* Check for no more waiters: */
if (TAILQ_FIRST(&(*cond)->c_queue) == NULL)
(*cond)->c_mutex = NULL;
/* Unlock the condition variable structure: */
_SPINUNLOCK(&(*cond)->lock);
} else {
/*
* Schedule the next thread and unlock
* the condition variable structure:
*/
_thread_kern_sched_state_unlock(PS_COND_WAIT,
&(*cond)->lock, __FILE__, __LINE__);
done = (seqno != (*cond)->c_seqno);
interrupted = curthread->interrupted;
/*
* Check if the wait was interrupted
* (canceled) or needs to be resumed
* after handling a signal.
*/
if (interrupted != 0) {
/*
* Lock the mutex and ignore any
* errors. Note that even
* though this thread may have
* been canceled, POSIX requires
* that the mutex be reacquired
* prior to cancellation.
*/
(void)_mutex_cv_lock(mutex);
} else {
/*
* Lock the condition variable
* while removing the thread.
*/
_SPINLOCK(&(*cond)->lock);
cond_queue_remove(*cond,
curthread);
/* Check for no more waiters: */
if (TAILQ_FIRST(&(*cond)->c_queue) == NULL)
(*cond)->c_mutex = NULL;
_SPINUNLOCK(&(*cond)->lock);
/* Lock the mutex: */
rval = _mutex_cv_lock(mutex);
/*
* Return ETIMEDOUT if the wait
* timed out and there wasn't an
* error locking the mutex:
*/
if ((curthread->timeout != 0)
&& rval == 0)
rval = ETIMEDOUT;
}
}
}
break;
/* Trap invalid condition variable types: */
default:
/* Unlock the condition variable structure: */
_SPINUNLOCK(&(*cond)->lock);
/* Return an invalid argument error: */
rval = EINVAL;
break;
}
if ((interrupted != 0) && (curthread->continuation != NULL))
curthread->continuation((void *) curthread);
} while ((done == 0) && (rval == 0));
_thread_leave_cancellation_point();
/* Return the completion status: */
return (rval);
}
ssize_t
read(int fd, void *buf, size_t nbytes)
{
struct pthread *curthread = _get_curthread();
ssize_t ret;
int type;
/* This is a cancellation point: */
_thread_enter_cancellation_point();
/* POSIX says to do just this: */
if (nbytes == 0)
ret = 0;
/* Lock the file descriptor for read: */
else if ((ret = _FD_LOCK(fd, FD_READ, NULL)) == 0) {
/* Get the read/write mode type: */
type = _thread_fd_table[fd]->status_flags->flags & O_ACCMODE;
/* Check if the file is not open for read: */
if (type != O_RDONLY && type != O_RDWR) {
/* File is not open for read: */
errno = EBADF;
ret = -1;
}
/* Perform a non-blocking read syscall: */
else while ((ret = _thread_sys_read(fd, buf, nbytes)) < 0) {
if ((_thread_fd_table[fd]->status_flags->flags & O_NONBLOCK) == 0 &&
(errno == EWOULDBLOCK || errno == EAGAIN)) {
curthread->data.fd.fd = fd;
_thread_kern_set_timeout(NULL);
/* Reset the interrupted operation flag: */
curthread->interrupted = 0;
curthread->closing_fd = 0;
_thread_kern_sched_state(PS_FDR_WAIT,
__FILE__, __LINE__);
/*
* Check if the operation was
* interrupted by a signal or
* a closing fd.
*/
if (curthread->interrupted) {
errno = EINTR;
ret = -1;
break;
} else if (curthread->closing_fd) {
errno = EBADF;
ret = -1;
break;
}
} else {
break;
}
}
_FD_UNLOCK(fd, FD_READ);
}
/* No longer in a cancellation point: */
_thread_leave_cancellation_point();
return (ret);
}
int
closefrom(int fd)
{
int ret = 0;
int safe_fd;
int lock_fd;
int *flags;
_thread_enter_cancellation_point();
if (fd < 0 || fd >= _thread_max_fdtsize) {
errno = EBADF;
ret = -1;
} else {
safe_fd = _thread_kern_pipe[0] > _thread_kern_pipe[1] ?
_thread_kern_pipe[0] : _thread_kern_pipe[1];
/*
* close individual files until we get past the pipe
* fds. Attempting to close a pipe fd is a no-op.
*/
for (safe_fd++; fd < safe_fd; fd++)
close(fd);
flags = calloc((size_t)_thread_max_fdtsize, sizeof *flags);
if (flags == NULL) {
/* use calloc errno */
ret = -1;
} else {
/* Lock and record all fd entries */
for (lock_fd = fd; lock_fd < _thread_max_fdtsize; lock_fd++) {
if (_thread_fd_table[lock_fd] != NULL &&
_thread_fd_table[lock_fd]->state != FD_ENTRY_CLOSED) {
ret = _FD_LOCK(lock_fd, FD_RDWR_CLOSE, NULL);
if (ret != -1)
flags[lock_fd] = 1;
else
break;
}
}
if (ret != -1) {
/*
* Close the entries and reset the non-bocking
* flag when needed.
*/
for (lock_fd = fd; lock_fd < _thread_max_fdtsize; lock_fd++) {
if (flags[lock_fd] != 0) {
_thread_fd_entry_close(lock_fd);
}
}
/*
* Now let the system do its thing. It is not practical
* to try to prevent races with other threads that can
* create new file descriptors. We just have to assume
* the application is well behaved when using closefrom.
*/
ret = _thread_sys_closefrom(fd);
}
/*
* Unlock any locked entries.
*/
for (lock_fd = fd; lock_fd < _thread_max_fdtsize; lock_fd++) {
if (flags[lock_fd] != 0) {
_FD_UNLOCK(lock_fd, FD_RDWR_CLOSE);
}
}
free(flags);
}
}
_thread_leave_cancellation_point();
return (ret);
}
ssize_t
writev(int fd, const struct iovec * iov, int iovcnt)
{
struct pthread *curthread = _get_curthread();
int blocking;
int idx = 0;
int type;
ssize_t num = 0;
size_t cnt;
ssize_t n;
ssize_t ret;
struct iovec liov[20];
struct iovec *p_iov = liov;
/* This is a cancellation point: */
_thread_enter_cancellation_point();
/* Check if the array size exceeds to compiled in size: */
if (iovcnt > (int) (sizeof(liov) / sizeof(struct iovec))) {
/* Allocate memory for the local array: */
if ((p_iov = (struct iovec *)
malloc((size_t)iovcnt * sizeof(struct iovec))) == NULL) {
/* Insufficient memory: */
errno = ENOMEM;
_thread_leave_cancellation_point();
return (-1);
}
} else if (iovcnt <= 0) {
errno = EINVAL;
_thread_leave_cancellation_point();
return (-1);
}
/* Copy the caller's array so that it can be modified locally: */
memcpy(p_iov,iov,(size_t)iovcnt * sizeof(struct iovec));
/* Lock the file descriptor for write: */
if ((ret = _FD_LOCK(fd, FD_WRITE, NULL)) == 0) {
/* Get the read/write mode type: */
type = _thread_fd_table[fd]->status_flags->flags & O_ACCMODE;
/* Check if the file is not open for write: */
if (type != O_WRONLY && type != O_RDWR) {
/* File is not open for write: */
errno = EBADF;
_FD_UNLOCK(fd, FD_WRITE);
if (p_iov != liov)
free(p_iov);
_thread_leave_cancellation_point();
return (-1);
}
/* Check if file operations are to block */
blocking = ((_thread_fd_table[fd]->status_flags->flags & O_NONBLOCK) == 0);
/*
* Loop while no error occurs and until the expected number
* of bytes are written if performing a blocking write:
*/
while (ret == 0) {
/* Perform a non-blocking write syscall: */
n = _thread_sys_writev(fd, &p_iov[idx], iovcnt - idx);
/* Check if one or more bytes were written: */
if (n > 0) {
/*
* Keep a count of the number of bytes
* written:
*/
num += n;
/*
* Enter a loop to check if a short write
* occurred and move the index to the
* array entry where the short write
* ended:
*/
cnt = (size_t)n;
while (cnt > 0 && idx < iovcnt) {
/*
* If the residual count exceeds
* the size of this vector, then
* it was completely written:
*/
if (cnt >= p_iov[idx].iov_len)
/*
* Decrement the residual
* count and increment the
* index to the next array
* entry:
*/
cnt -= p_iov[idx++].iov_len;
else {
/*
* This entry was only
* partially written, so
* adjust it's length
* and base pointer ready
* for the next write:
*/
p_iov[idx].iov_len -= cnt;
p_iov[idx].iov_base =
(char *)p_iov[idx].iov_base
+ (ptrdiff_t)cnt;
cnt = 0;
}
}
} else if (n == 0) {
/*
* Avoid an infinite loop if the last iov_len is
* 0.
*/
while (idx < iovcnt && p_iov[idx].iov_len == 0)
idx++;
if (idx == iovcnt) {
ret = num;
break;
}
}
/*
* If performing a blocking write, check if the
* write would have blocked or if some bytes
* were written but there are still more to
* write:
*/
if (blocking && ((n < 0 && (errno == EWOULDBLOCK ||
errno == EAGAIN)) || (n >= 0 && idx < iovcnt))) {
curthread->data.fd.fd = fd;
_thread_kern_set_timeout(NULL);
/* Reset the interrupted operation flag: */
curthread->interrupted = 0;
curthread->closing_fd = 0;
_thread_kern_sched_state(PS_FDW_WAIT,
__FILE__, __LINE__);
/*
* Check if the operation was
* interrupted by a signal
*/
if (curthread->interrupted || curthread->closing_fd) {
if (num > 0) {
/* Return partial success: */
ret = num;
} else {
/* Return an error: */
if (curthread->closing_fd)
errno = EBADF;
else
errno = EINTR;
ret = -1;
}
}
/*
* If performing a non-blocking write,
* just return whatever the write syscall did:
*/
} else if (!blocking) {
/* A non-blocking call might return zero: */
ret = n;
break;
/*
* If there was an error, return partial success
* (if any bytes were written) or else the error:
*/
} else if (n < 0) {
if (num > 0)
ret = num;
else
ret = n;
/* Check if the write has completed: */
} else if (idx == iovcnt)
/* Return the number of bytes written: */
ret = num;
}
_FD_UNLOCK(fd, FD_WRITE);
}
/* If memory was allocated for the array, free it: */
if (p_iov != liov)
free(p_iov);
/* No longer in a cancellation point: */
_thread_leave_cancellation_point();
return (ret);
}
int
connect(int fd, const struct sockaddr * name, socklen_t namelen)
{
struct pthread *curthread = _get_curthread();
struct sockaddr tmpname;
socklen_t errnolen, tmpnamelen;
int ret;
/* This is a cancellation point: */
_thread_enter_cancellation_point();
if ((ret = _FD_LOCK(fd, FD_RDWR, NULL)) == 0) {
if ((ret = _thread_sys_connect(fd, name, namelen)) < 0) {
if (!(_thread_fd_table[fd]->status_flags->flags & O_NONBLOCK) &&
((errno == EWOULDBLOCK) || (errno == EINPROGRESS) ||
(errno == EALREADY) || (errno == EAGAIN))) {
curthread->data.fd.fd = fd;
/* Reset the interrupted operation flag: */
curthread->interrupted = 0;
curthread->closing_fd = 0;
/* Set the timeout: */
_thread_kern_set_timeout(NULL);
_thread_kern_sched_state(PS_FDW_WAIT, __FILE__, __LINE__);
/*
* Check if the operation was
* interrupted by a signal or
* a closing fd.
*/
if (curthread->interrupted) {
errno = EINTR;
ret = -1;
} else if (curthread->closing_fd) {
errno = EBADF;
ret = -1;
} else {
tmpnamelen = sizeof(tmpname);
/* 0 now lets see if it really worked */
if (((ret = _thread_sys_getpeername(fd, &tmpname, &tmpnamelen)) < 0) &&
(errno == ENOTCONN)) {
/*
* Get the error, this function
* should not fail
*/
errnolen = sizeof(errno);
_thread_sys_getsockopt(fd, SOL_SOCKET, SO_ERROR, &errno, &errnolen);
}
}
} else {
ret = -1;
}
}
_FD_UNLOCK(fd, FD_RDWR);
}
/* No longer in a cancellation point: */
_thread_leave_cancellation_point();
return (ret);
}
int
accept(int fd, struct sockaddr * name, socklen_t *namelen)
{
struct pthread *curthread = _get_curthread();
int ret;
int newfd;
enum fd_entry_mode init_mode;
/* This is a cancellation point: */
_thread_enter_cancellation_point();
/* Lock the file descriptor: */
if ((ret = _FD_LOCK(fd, FD_RDWR, NULL)) == 0) {
/* Enter a loop to wait for a connection request: */
while ((ret = _thread_sys_accept(fd, name, namelen)) < 0) {
/* Check if the socket is to block: */
if ((_thread_fd_table[fd]->status_flags->flags & O_NONBLOCK) == 0 &&
(errno == EWOULDBLOCK || errno == EAGAIN)) {
/* Save the socket file descriptor: */
curthread->data.fd.fd = fd;
curthread->data.fd.fname = __FILE__;
curthread->data.fd.branch = __LINE__;
/* Set the timeout: */
_thread_kern_set_timeout(NULL);
curthread->interrupted = 0;
curthread->closing_fd = 0;
/* Schedule the next thread: */
_thread_kern_sched_state(PS_FDR_WAIT, __FILE__,
__LINE__);
/* Check if the wait was interrupted: */
if (curthread->interrupted) {
/* Return an error status: */
errno = EINTR;
ret = -1;
break;
} else if (curthread->closing_fd) {
/* Return an error status: */
errno = EBADF;
ret = -1;
break;
}
} else {
/*
* Another error has occurred, so exit the
* loop here:
*/
break;
}
}
/*
* If no errors initialize the file descriptor table
* for the new socket. If the client's view of the
* status_flags for fd is blocking, then force newfd
* to be viewed as blocking too.
*/
if (ret != -1) {
newfd = ret;
if ((_thread_fd_table[fd]->status_flags->flags & O_NONBLOCK) == 0)
init_mode = FD_INIT_BLOCKING;
else
init_mode = FD_INIT_NEW;
if((ret = _thread_fd_table_init(newfd, init_mode, NULL)) != -1)
ret = newfd;
else {
/* quitely close the fd */
_thread_sys_close(ret);
}
}
/* Unlock the file descriptor: */
_FD_UNLOCK(fd, FD_RDWR);
}
/* No longer in a cancellation point: */
_thread_leave_cancellation_point();
/* Return the socket file descriptor or -1 on error: */
return (ret);
}