Re: Seg fault on "echo ~nosuchuser"

From	Chet Ramey <chet.ramey@case.edu>
Newsgroups	gnu.bash.bug
Subject	Re: Seg fault on "echo ~nosuchuser"
Date	2020-06-02 08:56 -0400
Organization	ITS, Case Western Reserve University
Message-ID	<mailman.974.1591102629.2541.bug-bash@gnu.org> (permalink)
References	(3 earlier) <81924c39-6cde-617b-c5a7-5e3605e0c3b8@case.edu> <CAAHpriP=25drsYLtrOti2-fYEFjNpm76sAVskiVxUdthEB0vjw@mail.gmail.com> <6e7b8282-c41a-15b8-0dbf-9f811f6c3eee@case.edu> <1591055912.1057.9.camel@16bits.net> <d816520f-b4b2-aa5c-e4bb-bfd1b94714d3@case.edu>

Show all headers | View raw

On 6/1/20 7:58 PM, Ángel wrote:
> On 2020-06-01 at 15:12 -0400, Chet Ramey wrote:
>> I finally found a case where 16-byte alignment for memory returned by
>> malloc() is required. But it's only on Linux systems that use systemd.
>> I bet it's trying to marshal arguments for IPC and uses instructions
>> that require 16-byte alignment.
>>
>> Thanks for your help verifying this.
>>
>> Chet
> 
> You mean that with systemd getpwnam() crashes if using a malloc() that
> returns addresses not 16-byte aligned?

Yes, that's what I mean. The only significant change between the bash
version that worked and the one(s) that did not is making the memory
the bash malloc returns 16-byte aligned.

Other systems not using glibc or systemd could not reproduce the issue.

> Cornercase, surely, but it seems like a bug in whatever is assuming such
> alignment. That's not even pointer-size alignment.

If systemd or its support libraries are using copy instructions that
require 16-byte alignment, though it's not documented, it's arguably not
a bug.

Chet

-- 
``The lyf so short, the craft so long to lerne.'' - Chaucer
		 ``Ars longa, vita brevis'' - Hippocrates
Chet Ramey, UTech, CWRU    chet@case.edu    http://tiswww.cwru.edu/~chet/

Back to gnu.bash.bug | Previous | Next | Find similar

Thread

Re: Seg fault on "echo ~nosuchuser" Chet Ramey <chet.ramey@case.edu> - 2020-06-02 08:56 -0400

csiph-web