| Message ID | 20241121111506.4717-1-gpdev@gpost.dk |
|---|---|
| Headers | show |
| Series | Fix to allow more correct isatty() | expand |
On Thursday, November 21, 2024 6:12:53 AM EST Gil Pedersen wrote: > This patch comes from an issue discovered in systemd, where it can fail to > restore a text TTY session after a GUI session is stopped when compiled > with musl libc. > > The specific systemd integration issue is currently resolved in the musl > master branch by closer aligning the implementation with glibc, but the > underlying isatty() implementation is still flawed since it can return 0 > (false) for something that is definitely a TTY. Essentially both musl > and glibc give up correctly handling this case on Linux due to > inadequate/buggy kernel APIs. > > Thus I am proposing this patch as a solution to fix the kernel APIs. An > alternative fix could be to add a new IOCTL to specifically handle this, > but that seems overkill. > This is pretty cool as it fixes the /dev/console-is-a-disconnected-/dev/ttyS0 issue I found this year https://github.com/systemd/systemd/pull/33690 and https://lore.kernel.org/all/2669238.7s5MMGUR32@nerdopolis2/ namely on vt-less kernels that now have /dev/console as /dev/ttyS0 instead of /dev/tty0 or a virtual device. So yeah, when you have hardware and you've been using vt-enabled kernels on it for years (where your /dev/ttyS0 doesn't even have an rs232 port, let alone connected to anything), and then when you upgrade to a vt-less kernel, sometime in the future, because of systemd's isatty() check failing, and refusing to write to /dev/console, Plymouth's new log display facility in turn gets far fewer logs from systemd, because of the ioctl failure in isatty() MY idea for this solution was that kernels turning off CONFIG_VT_CONSOLE could then turn on a CONFIG_NULL_TTY_CONSOLE that would get selected at the same time the VT console would have, so that if distros choose to do so, the virtual /dev/ttynull device to always work for User Space. (that way kernels that never had CONFIG_VT_CONSOLE turned on won't have a behavior change as well) (the one benefit to my idea is that Plymouth goes into text mode logging when the console device is /dev/ttyS0, so Desktop distros would need to add a new "plymouth.graphical" to their grub commands to get the splash to work again, however when the console device is /dev/ttynull it automatically assumes that the user wants a splash screen.) This is more specific to Plymouth though However testing this patch and it also works, and this won't need a new configuration change. > Link: https://github.com/systemd/systemd/pull/34039 > Link: https://gitlab.alpinelinux.org/alpine/aports/-/merge_requests/70711 > Link: https://www.openwall.com/lists/musl/2024/08/20/2 > Link: https://git.musl-libc.org/cgit/musl/commit/src/unistd/isatty.c?id=c94a0c16f08894ce3be6dafb0fe80baa77a6ff2a > Link: https://sourceware.org/bugzilla/show_bug.cgi?id=32103 > > Gil Pedersen (1): > tty: respond to TIOCGWINSZ when hung > > drivers/tty/tty_io.c | 29 ++++++++++++++++++++++------- > 1 file changed, 22 insertions(+), 7 deletions(-) > >