A domain without a console quietly dies soon after start, because we try to set /dev/null as a controlling TTY 2014-10-30 15:10:59.705+0000: 1: error : lxcContainerSetupFDs:283 : ioctl(TIOCSTTY) failed: Inappropriate ioctl for device

Report an error early instead of trying to start it. https://bugzilla.redhat.com/show_bug.cgi?id=1155410 --- src/lxc/lxc_container.c | 6 ++++-- src/lxc/lxc_process.c | 6 ++++++ 2 files changed, 10 insertions(+), 2 deletions(-) diff --git a/src/lxc/lxc_container.c b/src/lxc/lxc_container.c index f02b959..8aba3ba 100644 --- a/src/lxc/lxc_container.c +++ b/src/lxc/lxc_container.c @@ -2093,8 +2093,10 @@ static int lxcContainerChild(void *data) if (virAsprintf(&ttyPath, "%s/%s.devpts/%s", LXC_STATE_DIR, vmDef->name, tty) < 0) goto cleanup; - } else if (VIR_STRDUP(ttyPath, "/dev/null") < 0) { - goto cleanup; + } else { + virReportError(VIR_ERR_INTERNAL_ERROR, "%s", + _("At least one tty is required")); + goto cleanup; }

On 10/31/2014 11:53 AM, Martin Kletzander wrote: > On Fri, Oct 31, 2014 at 10:22:25AM +0100, Ján Tomko wrote: >> A domain without a console quietly dies soon after start, >> because we try to set /dev/null as a controlling TTY >> 2014-10-30 15:10:59.705+0000: 1: error : lxcContainerSetupFDs:283 : >> ioctl(TIOCSTTY) failed: Inappropriate ioctl for device >> > > s/TIOCSTTY/TIOCSCTTY/ would make it more greppable (is that even a > word?) and it's now true since you pushed your trivial fix for that ;) > I have amended the commit message. >> Report an error early instead of trying to start it. >> >> https://bugzilla.redhat.com/show_bug.cgi?id=1155410 >> --- >> src/lxc/lxc_container.c | 6 ++++-- >> src/lxc/lxc_process.c | 6 ++++++ >> 2 files changed, 10 insertions(+), 2 deletions(-) >> >> diff --git a/src/lxc/lxc_container.c b/src/lxc/lxc_container.c >> index f02b959..8aba3ba 100644 >> --- a/src/lxc/lxc_container.c >> +++ b/src/lxc/lxc_container.c >> @@ -2093,8 +2093,10 @@ static int lxcContainerChild(void *data) >> if (virAsprintf(&ttyPath, "%s/%s.devpts/%s", >> LXC_STATE_DIR, vmDef->name, tty) < 0) >> goto cleanup; >> - } else if (VIR_STRDUP(ttyPath, "/dev/null") < 0) { >> - goto cleanup; >> + } else { >> + virReportError(VIR_ERR_INTERNAL_ERROR, "%s", >> + _("At least one tty is required")); >> + goto cleanup; >> } > > How about we don't attach any tty and use the ioctl(TIOCSTTY) to steal > controlling terminal (or call open() with O_NOCTTY, or just don't call > ioctl() at all, I don't know what we normally use)? Then there would > be no controlling terminal, but there would still be a session leader, > wouldn't it? > I don't know if we want to run a container without a controlling terminal. Looking at git history, back when we only supported one console and none was specified, it seems we quietly opened one anyway and didn't report it in the XML. Alternatively, we could auto-add a console in the XML. > Anyway, this version looks OK too since it doesn't break any use case > (there is nothing to break now). ACK after release if you argue why > not to use the idea above. It's been broken long enough to make me think nobody cares about this functionality, but I could make it work if you think it's worthwhile.