> -----Original Message-----
> From: Jason Wang [mailto:jasow...@redhat.com]
> Sent: Monday, January 15, 2018 3:21 PM
> To: Zhoujian (jay) <jianjay.z...@huawei.com>; qemu-devel@nongnu.org
> Cc: m...@redhat.com; imamm...@redhat.com; Huangweidong (C)
> <weidong.hu...@huawei.com>; wangxin (U) <wangxinxin.w...@huawei.com>; Gonglei
> (Arei) <arei.gong...@huawei.com>
> Subject: Re: [PATCH] tap: close fd conditionally when error occured
>
>
>
> On 2018年01月12日 15:30, Jay Zhou wrote:
> > If netdev_add tap,id=net0,vhost=on failed in net_init_tap_one(), the
> > followed up device_add virtio-net-pci,netdev=net0 will fail too,
> > prints:
> >
> > TUNSETOFFLOAD ioctl() failed: Bad file descriptor TUNSETOFFLOAD
> > ioctl() failed: Bad file descriptor
> >
> > The reason is that the fd of tap is closed when error occured after
> > calling net_init_tap_one().
> >
> > I think the fd should be closed in these two case:
> > 1.tap_set_sndbuf() failed
> > 2.tap_set_sndbuf() succeeded but vhost failed to initialize with
> > vhostforce flag on
> > Meanwhile, the fd should not be closed just because vhost failed to
> > initialize but without vhostforce flag. So the followed up device_add
> > can fall back to userspace virtio successfully.
> >
> > Suggested-by: Michael S. Tsirkin <m...@redhat.com>
> > Suggested-by: Igor Mammedov <imamm...@redhat.com>
> > Suggested-by: Jason Wang <jasow...@redhat.com>
> > Signed-off-by: Jay Zhou <jianjay.z...@huawei.com>
> > ---
> > net/tap.c | 18 +++++++++++++-----
> > 1 file changed, 13 insertions(+), 5 deletions(-)
> >
> > diff --git a/net/tap.c b/net/tap.c
> > index 979e622..3ed72eb 100644
> > --- a/net/tap.c
> > +++ b/net/tap.c
> > @@ -651,6 +651,9 @@ static void net_init_tap_one(const NetdevTapOptions
> *tap, NetClientState *peer,
> > tap_set_sndbuf(s->fd, tap, &err);
> > if (err) {
> > error_propagate(errp, err);
> > + if (!tap->has_fd && !tap->has_fds) {
> > + close(fd);
> > + }
> > return;
> > }
> >
> > @@ -687,14 +690,14 @@ static void net_init_tap_one(const NetdevTapOptions
> *tap, NetClientState *peer,
> > vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err);
> > if (vhostfd == -1) {
> > error_propagate(errp, err);
> > - return;
> > + goto cleanup;
> > }
> > } else {
> > vhostfd = open("/dev/vhost-net", O_RDWR);
> > if (vhostfd < 0) {
> > error_setg_errno(errp, errno,
> > "tap: open vhost char device failed");
> > - return;
> > + goto cleanup;
> > }
> > fcntl(vhostfd, F_SETFL, O_NONBLOCK);
> > }
> > @@ -704,11 +707,18 @@ static void net_init_tap_one(const NetdevTapOptions
> *tap, NetClientState *peer,
> > if (!s->vhost_net) {
> > error_setg(errp,
> > "vhost-net requested but could not be
> > initialized");
>
> So error_setg() is not appropriate here consider it was not trated as an
> error. We probably just need some warning.
Should it be warning too if vhostforce flag on?
>
> > - return;
> > + goto cleanup;
> > }
> > } else if (vhostfdname) {
> > error_setg(errp, "vhostfd(s)= is not valid without vhost");
> > }
> > +
> > +cleanup:
> > + if (!tap->has_fd && !tap->has_fds && tap->has_vhostforce &&
> > + tap->vhostforce) {
> > + close(fd);
>
> I would still let caller to decide whether or not to close the fd.
How about moving tap_set_sndbuf() out of net_init_tap_one()? So the caller
could decide whether or not close the fd easily(The reason is explained in
the commit message). Any better suggestion?
Regards,
Jay
>
> Thanks
>
> > + }
> > + return;
> > }
> >
> > static int get_fds(char *str, char *fds[], int max) @@ -877,7 +887,6
> > @@ free_fail:
> > vnet_hdr, fd, &err);
> > if (err) {
> > error_propagate(errp, err);
> > - close(fd);
> > return -1;
> > }
> > } else {
> > @@ -916,7 +925,6 @@ free_fail:
> > vhostfdname, vnet_hdr, fd, &err);
> > if (err) {
> > error_propagate(errp, err);
> > - close(fd);
> > return -1;
> > }
> > }
