On Fri, Sep 14, 2018 at 09:38:52PM +0200, Arnd Bergmann wrote: > On Fri, Sep 14, 2018 at 8:17 PM gregkh <gre...@linuxfoundation.org> wrote: > > > > On Fri, Sep 14, 2018 at 05:15:52PM +0200, Arnd Bergmann wrote: > > > On Thu, Sep 13, 2018 at 4:40 AM Al Viro <v...@zeniv.linux.org.uk> wrote: > > > > + case TCSETX: > > > + case TCSETXF: > > > + case TCSETXW: > > > + case TIOCGETC: > > > + case TIOCGETP: > > > + case TIOCGPTPEER: > > > + case TIOCSERCONFIG: > > > + case TIOCSETC: > > > + case TIOCSETN: > > > + case TIOCSETP: > > > + case TIOCVHANGUP: > > > > > > Should these all be added in a stable backport patch? > > > > Why? What is being "fixed" here that anyone has noticed before that is > > causing problems in those old kernels? If no one has complained about > > the lack of compat fixups, well... :) > > Most of the individual ones don't matter: > > TCSETX/TCSETXF/TCSETXW/TCGETX never did anything, > we don't care about them. > > TIOCSERCONFIG has always been broken in compat mode, > not sure who calls it in practice. This is 'setserial autoconfig'. > > TIOCGETC/TIOCGETP/TIOCSETC/TIOCSETN/TIOCSETP > are only defined on alpha, mips, powerpc and sparc to start > with. They seem to only be there for compatibility with some > Unix version for those architectures but were never part of > the normal Linux API or the compat set, so I guess they > also don't matter. > > The two that I think we may care about are: > > TIOCGPTPEER is a fairly recent regression from commit > 311fc65c9fb9 ("pty: Repair TIOCGPTPEER") in 4.14 after it was > added in 4.13. > > TIOCVHANGUP was introduced to be used by systemd in > commit 3c95c985fa91 ("tty: add TIOCVHANGUP to allow clean tty > shutdown of all ttys"). > https://raspberrypi.stackexchange.com/questions/54143 > has a bug report noting a problem with Raspbian systemd > that was related to TIOCVHANGUP. I don't think it's the > missing compat line that caused it, but the same system > would try to run TIOCVHANGUP in compat mode on > Raspberry Pi3 when using a 64-bit kernel with raspbian, > so there are definitely callers.
Ah, ok, that makes sense, I forgot about the new ioctls and didn't realize it was a regression. greg k-h