Eric Blake <eblake(a)redhat.com> wrote on 09/24/2010 06:16:35 PM:
On 09/24/2010 02:22 PM, Stefan Berger wrote:
> I just tried the TCK test without and with double-escaping in libvirtd
> and double-escaping does seem to be necessary otherwise `ls` and $(ls)
> do get executed and their results end up in the comment. The spaces
> are preserved, though, so I can revert the change to IFS.
Hmm.
> "res=`eval \"$cmd\"" CMD_SEPARATOR
> + virBufferVSprintf(buf,
> + " -m comment --comment \\\"%s\\\"",
> + cmt);
Thinking about it more:
[...]
My suggestion is to assign cmd using '' rather than "" (fewer things to
quote), as well as moving the eval outside of the `` (so it becomes
obvious which \ are interpreted by eval rather than by ``:
cmd='iptables -m comment --comment '\''user $comment'\''
eval res=\`"$cmd"\`
res=`iptables -m comment --comment 'user $comment'`
And the nice part of that is the implementation:
virBufferVSprintf(buf, " -m comment --comment '%s'",
escapeSingleQuotes(user_comment));
virBufferVSprintf(cmd, "cmd='%s'\nres=\\`\"$cmd\"\\`",
escapeSingleQuotes(buf));
Also I followed this. I had to write it like this here to reflect what you
wrote further above:
virBufferVSprintf(buf, " -m comment --comment
'\''%s'\''",
shellEscapeString(user_comment));
and shellEscapeString() needs to escape ' as well as `, otherwise I can
execute commands.
Thanks for the help.
On further thought, gnulib might be doing:
#define strchr rpl_strchr
on platforms where strchr is broken, so using #undef strchr is too
risky. So I'd recommend sticking with (strchr)(a, b), which still works
if gnulib has to replace a broken strchr.
Ok. This also works. Wished they left a note in the man pages about
this...
Stefan