Hey Joe,
Which screen version are you using?
Try compiling from master, the issue should be solved there.
-David
El sáb., 6 jun. 2020 0:39, Joe Yeung <joe.a.ye...@gmail.com> escribió:
> Hi,
>
> I am using screen to connect to /dev/pts/X which is mapped to a VM's
> serial port using QEMU.
> The first time I connect to it, as a non-privilege user, the connection
> went through.
>
> However, if I connect to it again, screen immediately exists with the
> "screen is terminating" message. The only way around it is sudo.
>
> I notice also that if the first connection is with sudo, subsequent
> connections as a non-privilege user also failed.
>
> It is as if the /dev/pts/X port was put into a wrong state after initial
> connection closes.
>
> So I re-compiled with debug turned on which led me to this function where
> the difference is made between the initial and the subsequent connection.
>
> int secopen(char *name, int flags, int mode)
> {
> int fd;
>
> xseteuid(real_uid);
> xsetegid(real_gid);
> fd = open(name, flags, mode);
> xseteuid(eff_uid);
> xsetegid(eff_gid);
> return fd;
> }
>
>
> The very first time I connection, open() succeeded, but subsequent ones
> always return -1 with the errno of EBUSY, but I don't see anything
> holding the file in lsof.
> The open() API was called with the same arguments every time, so why the
> different result?
> I also tried using strace, but I don't see differences between the first
> working connection and the subsequent failing connections.
>
> One thing I couldn't understand is why we seteuid to real uid before
> opening /dev/pts/X and reset it to effective uid afterwards? Should it be
> the other way around? Set to effective uid first then reset to the real uid
> ?? However, this logic had been there for 5 years.. The secfopen() API also
> has similar logic - set to real uid and reset to effective uid.
>
> I took a stab and swap the seteuid order and it worked.
>
> Is there a reason why /dev/pts/X can only be opened by a non-privilege
> user once?
> Am I running into a 5 years old bug?
>
> thanks!
>
