On Tue, Sep 6, 2016 at 9:43 PM, Peter Eisentraut
<peter.eisentraut@2ndquadrant.com> wrote:
> On 8/11/16 9:12 PM, Michael Paquier wrote:
>> Note that pg_dump[all] and pg_upgrade already have safeguards against
>> those things per the same routines putting quotes for execution as
>> commands into psql and shell. So attached is a patch to implement this
>> restriction in the backend, and I am adding that to the next CF for
>> 10.0. Attached is as well a script able to trigger those errors.
>
> After further review, I have my doubts about this approach.
>
> Everything that is using appendShellString() is now going to reject LF
> and CR characters, but there is no systematic way by which this is
> managed, enforced, or documented. It happens that right now most of the
> affected cases are user and database names, but there are others. For
> example, you cannot anymore install PostgreSQL into a path containing
> LF/CR, because initdb will fail when it composes the pg_ctl command line
> to print out. Also, initdb will fail if the data directory name
> contains LF/CR, but it creates the directory nonetheless. (Apparently,
> it doesn't even clean it up.) But for example pg_ctl and pg_basebackup
> and postgres itself handle all of that just fine. This is a slowly
> growing mess.
>
> I think the way forward here, if any, is to work on removing these
> restrictions, not to keep sprinkling them around.
If we were talking about pathnames containing spaces, I would agree,
but I've never heard of a legitimate pathname containing CR or LF. I
can't see us losing much by refusing to allow such pathnames, except
for security holes.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company