Re: [PATCH 20/52] login: Add 64-bit time support to utmp/utmpx

[Florian Weimer via Libc-alpha <libc-alpha@sourceware.org>]

* Adhemerval Zanella via Libc-alpha:

> The new struct has the same size for 32-bit and 64-bit architecture with
> two main differences compared to the 32-bit time one:

I think we should deprecate the interfaces instead.

Due to a local denial-of-service vulnerability (bug 24492), we would
have to switch to a daemon.

However, the wtmp logging is rather useless these days because the main
key is ut_id, the “terminal name suffix” (so presumably S0 for the first
serial console).  But most terminals are pseudo-terminals today, and it
is fairly meaningless which user was logged on on which pseudo-terminal.
glibc offers an additional lookup procedure with unique ut_line keys as
an extension.  But even if meaningful keys are used there (such as host
names), another issue appears: the interface is designed in such a way
that the set of lookup keys must be very small because old entries are
never expired, only overwritten if the lookup key is used again.  The
challenge with the present interface is that applications need to come
up with a lookup key that is unique, meaningful, and will be reused
quickly in case the session terminates.  Historically, it was possible
to use the PID (and vfsftpd does that), but the PID limit is gone on
some distributions, and the PID in the index leads to essentially
unbounded utmp file growth.

The interface was fine when a system had a few fixed terminal lines that
weren't equivalent.  But I think it's mostly useless today.

Thanks,
Florian