int
pipe1(struct lwp *l, register_t *retval, int flags)
{
file_t *rf, *wf;
struct socket *rso, *wso;
int fd, error;
proc_t *p;
if (flags & ~(O_CLOEXEC|O_NONBLOCK|O_NOSIGPIPE))
return EINVAL;
p = curproc;
if ((error = socreate(AF_LOCAL, &rso, SOCK_STREAM, 0, l, NULL)) != 0)
return error;
if ((error = socreate(AF_LOCAL, &wso, SOCK_STREAM, 0, l, rso)) != 0)
goto free1;
/* remember this socket pair implements a pipe */
wso->so_state |= SS_ISAPIPE;
rso->so_state |= SS_ISAPIPE;
if ((error = fd_allocfile(&rf, &fd)) != 0)
goto free2;
retval[0] = fd;
rf->f_flag = FREAD | flags;
rf->f_type = DTYPE_SOCKET;
rf->f_ops = &socketops;
rf->f_socket = rso;
if ((error = fd_allocfile(&wf, &fd)) != 0)
goto free3;
wf->f_flag = FWRITE | flags;
wf->f_type = DTYPE_SOCKET;
wf->f_ops = &socketops;
wf->f_socket = wso;
retval[1] = fd;
solock(wso);
error = unp_connect2(wso, rso);
sounlock(wso);
if (error != 0)
goto free4;
fd_affix(p, wf, (int)retval[1]);
fd_affix(p, rf, (int)retval[0]);
return (0);
free4:
fd_abort(p, wf, (int)retval[1]);
free3:
fd_abort(p, rf, (int)retval[0]);
free2:
(void)soclose(wso);
free1:
(void)soclose(rso);
return error;
}
/* open the xenevt device; this is where we clone */
int
xenevtopen(dev_t dev, int flags, int mode, struct lwp *l)
{
struct xenevt_d *d;
struct file *fp;
int fd, error;
switch(minor(dev)) {
case DEV_EVT:
/* falloc() will use the descriptor for us. */
if ((error = fd_allocfile(&fp, &fd)) != 0)
return error;
d = malloc(sizeof(*d), M_DEVBUF, M_WAITOK | M_ZERO);
d->ci = &cpu_info_primary;
mutex_init(&d->lock, MUTEX_DEFAULT, IPL_HIGH);
cv_init(&d->cv, "xenevt");
selinit(&d->sel);
return fd_clone(fp, fd, flags, &xenevt_fileops, d);
case DEV_XSD:
/* no clone for /dev/xsd_kva */
return (0);
default:
break;
}
return ENODEV;
}
static int
makesocket(struct lwp *l, file_t **fp, int *fd, int flags, int type,
int domain, int proto, struct socket *soo)
{
struct socket *so;
int error;
if ((error = socreate(domain, &so, type, proto, l, soo)) != 0) {
return error;
}
if (flags & SOCK_NONBLOCK) {
so->so_state |= SS_NBIO;
}
if ((error = fd_allocfile(fp, fd)) != 0) {
soclose(so);
return error;
}
fd_set_exclose(l, *fd, (flags & SOCK_CLOEXEC) != 0);
(*fp)->f_flag = FREAD|FWRITE|
((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
(*fp)->f_type = DTYPE_SOCKET;
(*fp)->f_ops = &socketops;
(*fp)->f_socket = so;
return 0;
}
static int
pty_alloc_slave(struct lwp *l, int *fd, dev_t dev, struct mount *mp)
{
int error;
struct file *fp;
struct vnode *vp;
/* Grab a filedescriptor for the slave */
if ((error = fd_allocfile(&fp, fd)) != 0) {
DPRINTF(("fd_allocfile %d\n", error));
return error;
}
if (ptm == NULL) {
error = EOPNOTSUPP;
goto bad;
}
if ((error = (*ptm->allocvp)(mp, l, &vp, dev, 't')) != 0)
goto bad;
if ((error = pty_vn_open(vp, l)) != 0)
goto bad;
fp->f_flag = FREAD|FWRITE;
fp->f_type = DTYPE_VNODE;
fp->f_ops = &vnops;
fp->f_data = vp;
VOP_UNLOCK(vp);
fd_affix(curproc, fp, *fd);
return 0;
bad:
fd_abort(curproc, fp, *fd);
return error;
}
static int
xen_dev_open(dev_t dev, int flags, int mode, struct lwp *l)
{
const struct xen_dev_info *xdinfo;
int fd, err;
struct file *fp;
void *fdata;
DPRINTF(("xen devsw: opening minor=%lu\n", (unsigned long)minor(dev)));
if (minor(dev) < 0 || minor(dev) >= NUM_DEV_INFOS)
return ENODEV;
xdinfo = &devs[minor(dev)];
if (!xdinfo->xd_open)
return ENODEV;
err = fd_allocfile(&fp, &fd);
if (err)
return err;
DPRINTF(("%s: opening...\n", xdinfo->path));
err = xdinfo->xd_open(fp, &fdata);
if (err) {
fd_abort(curproc, fp, fd);
return err;
}
DPRINTF(("%s: opened, fd_clone\n", xdinfo->path));
return fd_clone(fp, fd, flags, xdinfo->fo, fdata);
}
int sys_tunopen(struct lwp *l, const struct sys_tunopen_args *uap, register_t *retval)
{
int error;
struct file* fp;
int indx;
error = ChkSpaceBstrR(SCARG(uap, path), 0);
if ( error < E_OK ) {
return EFAULT;
}
if ((error = fd_allocfile(&fp, &indx)) != 0)
return (error);
if ((error = tkn_tun_open(SCARG(uap, path), SCARG(uap, oflag), fp)) != 0) {
fd_abort(l->l_proc, fp, indx);
if ((error == EDUPFD || error == EMOVEFD) &&
l->l_dupfd >= 0 && /* XXX from fdopen */
(error =
fd_dupopen(l->l_dupfd, &indx, SCARG(uap, oflag), error)) == 0) {
*retval = indx;
return (0);
}
if (error == ERESTART)
error = EINTR;
return (error);
}
return ENOTSUP;
}
int
rndopen(dev_t dev, int flags, int fmt, struct lwp *l)
{
bool hard;
struct file *fp;
int fd;
int error;
switch (minor(dev)) {
case RND_DEV_URANDOM:
hard = false;
break;
case RND_DEV_RANDOM:
hard = true;
break;
default:
return ENXIO;
}
error = fd_allocfile(&fp, &fd);
if (error)
return error;
/*
* Allocate a context, but don't create a CPRNG yet -- do that
* lazily because it consumes entropy from the system entropy
* pool, which (currently) has the effect of depleting it and
* causing readers from /dev/random to block. If this is
* /dev/urandom and the process is about to send only short
* reads to it, then we will be using a per-CPU CPRNG anyway.
*/
struct rnd_ctx *const ctx = pool_cache_get(rnd_ctx_cache, PR_WAITOK);
ctx->rc_cprng = NULL;
ctx->rc_hard = hard;
error = fd_clone(fp, fd, flags, &rnd_fileops, ctx);
KASSERT(error == EMOVEFD);
return error;
}
static int
tap_dev_cloner(struct lwp *l)
{
struct tap_softc *sc;
file_t *fp;
int error, fd;
if ((error = fd_allocfile(&fp, &fd)) != 0)
return (error);
if ((sc = tap_clone_creator(-1)) == NULL) {
fd_abort(curproc, fp, fd);
return (ENXIO);
}
sc->sc_flags |= TAP_INUSE;
return fd_clone(fp, fd, FREAD|FWRITE, &tap_fileops,
(void *)(intptr_t)device_unit(sc->sc_dev));
}
/*
* dmoverioopen:
*
* Device switch open routine.
*/
int
dmoverioopen(dev_t dev, int flag, int mode, struct lwp *l)
{
struct dmio_state *ds;
struct file *fp;
int error, fd, s;
/* falloc() will use the descriptor for us. */
if ((error = fd_allocfile(&fp, &fd)) != 0)
return (error);
s = splsoftclock();
ds = pool_get(&dmio_state_pool, PR_WAITOK);
splx(s);
memset(ds, 0, sizeof(*ds));
simple_lock_init(&ds->ds_slock);
TAILQ_INIT(&ds->ds_pending);
TAILQ_INIT(&ds->ds_complete);
selinit(&ds->ds_selq);
return fd_clone(fp, fd, flag, &dmio_fileops, ds);
}
/*
* kqueue(2) system call.
*/
static int
kqueue1(struct lwp *l, int flags, register_t *retval)
{
struct kqueue *kq;
file_t *fp;
int fd, error;
if ((error = fd_allocfile(&fp, &fd)) != 0)
return error;
fp->f_flag = FREAD | FWRITE | (flags & (FNONBLOCK|FNOSIGPIPE));
fp->f_type = DTYPE_KQUEUE;
fp->f_ops = &kqueueops;
kq = kmem_zalloc(sizeof(*kq), KM_SLEEP);
mutex_init(&kq->kq_lock, MUTEX_DEFAULT, IPL_SCHED);
cv_init(&kq->kq_cv, "kqueue");
selinit(&kq->kq_sel);
TAILQ_INIT(&kq->kq_head);
fp->f_data = kq;
*retval = fd;
kq->kq_fdp = curlwp->l_fd;
fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0);
fd_affix(curproc, fp, fd);
return error;
}
int
svr4_netopen(dev_t dev, int flag, int mode, struct lwp *l)
{
int type, protocol;
int fd;
file_t *fp;
struct socket *so;
int error;
int family;
DPRINTF(("netopen("));
if (curlwp->l_dupfd >= 0) /* XXX */
return ENODEV;
switch (minor(dev)) {
case dev_udp:
family = AF_INET;
type = SOCK_DGRAM;
protocol = IPPROTO_UDP;
DPRINTF(("udp, "));
break;
case dev_tcp:
family = AF_INET;
type = SOCK_STREAM;
protocol = IPPROTO_TCP;
DPRINTF(("tcp, "));
break;
case dev_ip:
case dev_rawip:
family = AF_INET;
type = SOCK_RAW;
protocol = IPPROTO_IP;
DPRINTF(("ip, "));
break;
case dev_icmp:
family = AF_INET;
type = SOCK_RAW;
protocol = IPPROTO_ICMP;
DPRINTF(("icmp, "));
break;
case dev_unix_dgram:
family = AF_LOCAL;
type = SOCK_DGRAM;
protocol = 0;
DPRINTF(("unix-dgram, "));
break;
case dev_unix_stream:
case dev_unix_ord_stream:
family = AF_LOCAL;
type = SOCK_STREAM;
protocol = 0;
DPRINTF(("unix-stream, "));
break;
default:
DPRINTF(("%d);\n", minor(dev)));
return EOPNOTSUPP;
}
if ((error = fd_allocfile(&fp, &fd)) != 0)
return error;
if ((error = socreate(family, &so, type, protocol, l, NULL)) != 0) {
DPRINTF(("socreate error %d\n", error));
fd_abort(curproc, fp, fd);
return error;
}
error = fd_clone(fp, fd, flag, &svr4_netops, so);
fp->f_type = DTYPE_SOCKET;
(void)svr4_stream_get(fp);
DPRINTF(("ok);\n"));
return error;
}
int
do_sys_accept(struct lwp *l, int sock, struct mbuf **name,
register_t *new_sock, const sigset_t *mask, int flags, int clrflags)
{
file_t *fp, *fp2;
struct mbuf *nam;
int error, fd;
struct socket *so, *so2;
short wakeup_state = 0;
if ((fp = fd_getfile(sock)) == NULL)
return EBADF;
if (fp->f_type != DTYPE_SOCKET) {
fd_putfile(sock);
return ENOTSOCK;
}
if ((error = fd_allocfile(&fp2, &fd)) != 0) {
fd_putfile(sock);
return error;
}
nam = m_get(M_WAIT, MT_SONAME);
*new_sock = fd;
so = fp->f_socket;
solock(so);
if (__predict_false(mask))
sigsuspendsetup(l, mask);
if (!(so->so_proto->pr_flags & PR_LISTEN)) {
error = EOPNOTSUPP;
goto bad;
}
if ((so->so_options & SO_ACCEPTCONN) == 0) {
error = EINVAL;
goto bad;
}
if ((so->so_state & SS_NBIO) && so->so_qlen == 0) {
error = EWOULDBLOCK;
goto bad;
}
while (so->so_qlen == 0 && so->so_error == 0) {
if (so->so_state & SS_CANTRCVMORE) {
so->so_error = ECONNABORTED;
break;
}
if (wakeup_state & SS_RESTARTSYS) {
error = ERESTART;
goto bad;
}
error = sowait(so, true, 0);
if (error) {
goto bad;
}
wakeup_state = so->so_state;
}
if (so->so_error) {
error = so->so_error;
so->so_error = 0;
goto bad;
}
/* connection has been removed from the listen queue */
KNOTE(&so->so_rcv.sb_sel.sel_klist, NOTE_SUBMIT);
so2 = TAILQ_FIRST(&so->so_q);
if (soqremque(so2, 1) == 0)
panic("accept");
fp2->f_type = DTYPE_SOCKET;
fp2->f_flag = (fp->f_flag & ~clrflags) |
((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
fp2->f_ops = &socketops;
fp2->f_socket = so2;
if (fp2->f_flag & FNONBLOCK)
so2->so_state |= SS_NBIO;
else
so2->so_state &= ~SS_NBIO;
error = soaccept(so2, nam);
so2->so_cred = kauth_cred_dup(so->so_cred);
sounlock(so);
if (error) {
/* an error occurred, free the file descriptor and mbuf */
m_freem(nam);
mutex_enter(&fp2->f_lock);
fp2->f_count++;
mutex_exit(&fp2->f_lock);
closef(fp2);
fd_abort(curproc, NULL, fd);
} else {
fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
fd_affix(curproc, fp2, fd);
*name = nam;
}
fd_putfile(sock);
if (__predict_false(mask))
sigsuspendteardown(l);
return error;
bad:
sounlock(so);
m_freem(nam);
fd_putfile(sock);
fd_abort(curproc, fp2, fd);
if (__predict_false(mask))
sigsuspendteardown(l);
return error;
}
/* ARGSUSED */
int
sys_ktrace(struct lwp *l, const struct sys_ktrace_args *uap, register_t *retval)
{
/* {
syscallarg(const char *) fname;
syscallarg(int) ops;
syscallarg(int) facs;
syscallarg(int) pid;
} */
struct vnode *vp = NULL;
file_t *fp = NULL;
struct pathbuf *pb;
struct nameidata nd;
int error = 0;
int fd;
if (ktrenter(l))
return EAGAIN;
if (KTROP(SCARG(uap, ops)) != KTROP_CLEAR) {
/*
* an operation which requires a file argument.
*/
error = pathbuf_copyin(SCARG(uap, fname), &pb);
if (error) {
ktrexit(l);
return (error);
}
NDINIT(&nd, LOOKUP, FOLLOW, pb);
if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
pathbuf_destroy(pb);
ktrexit(l);
return (error);
}
vp = nd.ni_vp;
pathbuf_destroy(pb);
VOP_UNLOCK(vp);
if (vp->v_type != VREG) {
vn_close(vp, FREAD|FWRITE, l->l_cred);
ktrexit(l);
return (EACCES);
}
/*
* This uses up a file descriptor slot in the
* tracing process for the duration of this syscall.
* This is not expected to be a problem.
*/
if ((error = fd_allocfile(&fp, &fd)) != 0) {
vn_close(vp, FWRITE, l->l_cred);
ktrexit(l);
return error;
}
fp->f_flag = FWRITE;
fp->f_type = DTYPE_VNODE;
fp->f_ops = &vnops;
fp->f_data = (void *)vp;
vp = NULL;
}
error = ktrace_common(l, SCARG(uap, ops), SCARG(uap, facs),
SCARG(uap, pid), &fp);
if (KTROP(SCARG(uap, ops)) != KTROP_CLEAR)
fd_abort(curproc, fp, fd);
return (error);
}
static int
pty_alloc_master(struct lwp *l, int *fd, dev_t *dev, struct mount *mp)
{
int error;
struct file *fp;
struct vnode *vp;
int md;
if ((error = fd_allocfile(&fp, fd)) != 0) {
DPRINTF(("fd_allocfile %d\n", error));
return error;
}
retry:
/* Find and open a free master pty. */
*dev = pty_getfree();
md = minor(*dev);
if ((error = pty_check(md)) != 0) {
DPRINTF(("pty_check %d\n", error));
goto bad;
}
if (ptm == NULL) {
DPRINTF(("no ptm\n"));
error = EOPNOTSUPP;
goto bad;
}
/*
* XXX Since PTYFS has now multiple instance support, if we mounted
* more than one PTYFS we must check here the ptyfs_used_tbl, to find
* out if the ptyfsnode is under the appropriate mount and skip the
* node if not, because the pty could has been released, but
* ptyfs_reclaim didn't get a chance to release the corresponding
* node other mount point yet.
*
* It's important to have only one mount point's ptyfsnode for each
* appropriate device in ptyfs_used_tbl, else we will have a security
* problem, because every entry will have access to this device.
*
* Also we will not have not efficient vnode and memory usage.
* You can test this by changing a_recycle from true to false
* in ptyfs_inactive.
*/
if ((error = (*ptm->allocvp)(mp, l, &vp, *dev, 'p')) != 0) {
DPRINTF(("pty_allocvp %d\n", error));
goto bad;
}
if ((error = pty_vn_open(vp, l)) != 0) {
DPRINTF(("pty_vn_open %d\n", error));
/*
* Check if the master open failed because we lost
* the race to grab it.
*/
if (error != EIO)
goto bad;
error = !pty_isfree(md, 1);
DPRINTF(("pty_isfree %d\n", error));
if (error)
goto retry;
else
goto bad;
}
fp->f_flag = FREAD|FWRITE;
fp->f_type = DTYPE_VNODE;
fp->f_ops = &vnops;
fp->f_data = vp;
VOP_UNLOCK(vp);
fd_affix(curproc, fp, *fd);
return 0;
bad:
fd_abort(curproc, fp, *fd);
return error;
}
