Re: N_HDLC line discipline: Race condition

[Dianne Skoll]

On Thu, 23 May 2024 10:01:51 +0200
Jiri Slaby <jirislaby@xxxxxxxxxx> wrote:

> The driver definitely behaves as described. If the ldisc is used on a
> real HW. ptys are a different story -- it's not guaranteed there.
> Does it make sense to use nhdlc on a pty pair? I believe not.

OK.  Well, the use case was as follows:  pppd has an option called `pty`
that makes it execute an arbitrary program, connect to its standard
input and output via a pty pair, and send and receive PPP frames over that
pty pair.

rp-pppoe (https://dianne.skoll.ca/projects/rp-pppoe/) includes a
program called `pppoe` designed to be on the other end of that pty
pair and receive/transmit the PPP frames via PPPoE.  pppd includes a
`sync` option, and pppoe a `-s` option that enable N_HDLC on the pty.
This lets pppoe just read/write a frame at a time without worrying
about PPP framing bytes or dealing with PPP escape characters, which
reduces the CPU overhead of pppoe.

Now, the Linux kernel has had built-in support for PPPoE for many
years, and I was thinking of dropping userspace PPPoE support, but I
heard from a user who wants to keep it.  ucLinux, it seems, does not
support dlopen(), so this user can't use pppd's `plugin` option to
load the kernel-mode PPPoE support module and has to keep using
user-space PPPoE.

Sync support was added to pppoe decades ago and either it worked well,
nobody used it, or nobody reported a bug until recently.  But anyway,
if the consensus is that N_HDLC shouldn't be used on a pty pair, I'm
fine with that.  Perhaps a comment in the source file and a note in
the N_HDLC documentation would be good, and then I'll just remove
support for `sync` and `-s` from pppoe, since it can't be guaranteed
to work correctly.

Regards,

Dianne.