On 15.07.2014 11:19, Karel Zak wrote: > On Tue, Jul 15, 2014 at 09:57:12AM +0200, Stef Walter wrote: >> On 15.07.2014 09:47, Karel Zak wrote: >>> Now when I think about it.. would be better to have this feature >>> optional? I think it's overkill to have it enabled on all machines. >> >> Sure. >> >>> The another story is the signal usage (and possible race with login(1) as >>> discussed at https://bugzilla.redhat.com/show_bug.cgi?id=1110763). >>> >>> What about for example: >>> >>> agetty --reload-trigger <file> >> >> As I said in the above bugzilla bug, I don't mind doing this ... but in >> my opinion unix signals are meant exactly for this use case, and are >> used by many (most?) other linx processes as a reload trigger. > > Yes, it's pretty common classic solution, but I think that in this > case it is not too robust solution. > >>> and use inotify for the <file>. I guess this way also allows to avoid >>> the ioctl() tty voodoo and you can use select() for the inotify and >>> tty file descriptors. >> >> Unfortunately that's not how it works. When in canonical mode, the tty >> file descriptor only becomes readable once a complete line has been >> entered. I've verified that select() does not return until <enter> is >> pressed. > > Ah.. good point, but with non-canonical mode it works as expected, > right? So, we can use fd with VMIN=1 and inotify fd for the select(). Yup. > Maybe the best would be to implement all within agetty, I mean add > a special command > > agetty --reload [<tty>] > > to reload another already running agetty process(es). So rather than > "killall agetty" you have to use "agetty --reload". > > Then all the implementation will be a private thing and only the "--reload" > command line option will be a public API. Okay. I'll work on this and keep you posted. > Note that we does not control code of all login(1) implementations > and we have --login-program <path> to start arbitrary random thing > from agetty, so our solution have to be generic rather than rely > on any login(1) change. Well you don't need to modify login(1) (or other --login-program's) at all. SIG_IGN is kept across an exec. Stef -- To unsubscribe from this list: send the line "unsubscribe util-linux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
