[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20181116213845.GM5150@brightrain.aerifal.cx>
Date: Fri, 16 Nov 2018 16:38:45 -0500
From: Rich Felker <dalias@...c.org>
To: musl@...ts.openwall.com, Paul Sykes <paulsykes@...il.com>
Subject: Re: optopt in getopt not being set, only when building with
OpenWrt
On Fri, Nov 16, 2018 at 10:11:00PM +0100, Szabolcs Nagy wrote:
> * Paul Sykes <paulsykes@...il.com> [2018-11-16 17:54:23 +0800]:
> > On my OpenWrt builds I found that the optopt variable was never set
> > when calling the function getopt. I tested the same code on my desktop
> > using the gcc wrapper and optopt was set as expected.
> >
> > Debugging showed that optopt was being set correctly in getopt.c
> > however the variable was not being set in my calling code.
> >
> > Adding optopt to the file 'dynamic.list' resolved the issue when
> > building with OpenWrt. I noticed that all the other extern variables
> > for getopt appear in dynamic.list except optopt.
Yes, this is exactly right. It seems to have been an oversight. I'll
push a fix for inclusion in 1.1.21, and distros using 1.1.20 should
apply it since this is a notable regression. However,
> yeah the dynamic list entry is needed so the address of the
> object internally in the libc is the same as in the main
> executable in case of copy relocation (only matters for
> dynamic linked executables and pie executable may work
> since that does not require copy relocs).
>
> all targets with copy relocs are affected, but only musl
> version 1.1.20 has the problem i think.
On distros using default-pie, the impact might be low, depending on
arch, gcc version, and CFLAGS. There's never a *need* to use copy
relocations with pie, and at one point gcc never did, but I think with
current tooling it's more complicated whether copy relocations get
used or not.
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
