On 26/05/2019 09:01, Pyry Kontio via lfs-dev wrote:
> Hi Pierre and Bruce, thanks for the reply.
>
> Let me compare some of the listings to try to pinpoint the differences.
>
> On Debian 9, I get:
>> # ls -l /dev/{,pts/}ptmx
>> lrwxrwxrwx 1 root root 8 May 26 05:55 /dev/ptmx -> pts/ptmx
>> crw-rw-rw- 1 root root 5, 2 May 26 06:23 /dev/pts/ptmx
>
> And if I create a new dev folder as in the LFS book, I get
>> # mkdir -pv /new_dev
>> mkdir: created directory '/new_dev'
>> # mount -v --bind /dev /new_dev
>> mount: /dev bound on /new_dev.
>> # ls -l /new_dev/ptmx
>> lrwxrwxrwx 1 root root 8 May 26 06:40 /new_dev/ptmx -> pts/ptmx
>> # ls -l /new_dev/pts
>> total 0
>> # mount -vt devpts devpts /new_dev/pts -o gid=5,mode=620
>> mount: devpts mounted on /new_dev/pts.
>> # ls -l /new_dev/pts
>> total 0
>> c--------- 1 root root 5, 2 May 26 06:43 ptmx
>
> I can see points of interest: the /dev/ptmx file is not symlinked in your
> listing, unlike in mine. Furthermore, the default /dev/pts/ptmx file in my
> distro has the permissions 0666, whereas a freshly mounted devpts
> filesystem has a ptmx file with zero permissions.
>
> Btw. these listings are from Debian 9 on Docker on MacOS. I think that
> from how the Docker on non-Linux systems works, the kernel is provided
> by the Docker runtime.
>> uname -a
>> Linux 7a69bd692285 4.9.125-linuxkit #1 SMP Fri Sep 7 08:20:28
>> UTC 2018 x86_64 GNU/Linux
Ah, there are a lot of reports of a different behavior of /dev/pts with
docker. When you first reported this issue, I found one, that I cannot find
again today (I don't remember which search line I entered), which was giving
some explanation.
Well, found another one you might find useful
https://github.com/docker/for-linux/issues/77
But surely there is something going on with docker+/dev/pts.
>
> I wonder if this brings any light into the situation.
> Do I understand correctly that an automatized LFS CI system is able
> to run the GCC tests on user nobody even when the ptmx file on the
> newly mounted devpts has zero permissions?
When built in a VM or directly on the host, yes. I'd say docker is the problem.
>
> If it helps to see the steps I'm taking in detail, I have a work-in-progress
> repo that contains a Docker container and scripts that perform all steps
> I'm taking here:
>
> https://gitlab.com/golddranks/lfs_docker
>
Please do not top post...
Pierre
>
>
> 2019年5月14日(火) 0:54 Bruce Dubbs via lfs-dev
> <[email protected]>:
>>
>> On 5/13/19 10:05 AM, Pierre Labastie via lfs-dev wrote:
>>> On 13/05/2019 15:50, Pyry Kontio via lfs-dev wrote:
>>>> Hi, and thanks for the wonderful book.
>>>>
>>>> I'm trying to follow the standard LFS track, but I ran problems with
>>>> the GCC test suite in Chapter 6
>>>> (http://www.linuxfromscratch.org/lfs/view/stable/chapter06/gcc.html)
>>>> -- I get a massive amount of failed results.
>>>>
>>>> I tried to report a bug using your tracker system, but even after
>>>> creating an account, I'm unable to create new tickets, so I'm sending
>>>> the report to this mailing list. Pardon me if it's the wrong one for
>>>> these kinds of reports.
>>>>
>>>> Looking into it a bit, the problem seems to be that the book suggests
>>>> running the test suite as user nobody:
>>>>
>>>>> su nobody -s /bin/bash -c "PATH=$PATH make -k check"
>>>> The problem becomes clear when running some failing test suite in
>>>> verbose mode:
>>>>
>>>>> runtest -v --tool gcc gcc.c-torture/compile/compile.exp
>>>> It can't create ptys. Double-checking:
>>>>
>>>>> expect -c "spawn ls"
>>>> ...fails although it succeeded when originally ran when building
>>>> Binutils.
>>>>
>>>> The problem seems to be that the file /dev/pts/ptmx that was created
>>>> when mounting devpts filesystem in 6.2
>>>> (http://www.linuxfromscratch.org/lfs/view/stable/chapter06/kernfs.html)
>>>> and is symlinked from /dev/ptmx on modern systems, has zero
>>>> permissions by default. This makes creating ptys with non-root
>>>> permissions fail. Some distros like Debian 9 set the system default
>>>> /dev/pts/ptmx to have permissions 0666. Indeed, setting it to have
>>>> those permissions will allow the nobody user to run the GCC test suite
>>>> successfully.
>>>>
>>>> Maybe an additional step could be added to chapter 6.2 to ensure that
>>>> /dev/pts/ptmx is set to have sensible permissions?
>>>
>>> Hi,
>>> You used the right list. I think you have to wait for approval from
>>> Bruce Dubbs for being able to report a bug.
>>
>> Correct. Needed to prevent spam. Better here or lfs-support. Most
>> reports (but not all) like this are invalid as we test by automating
>> builds by extracting the instructions from the book.
>>
>>> What is your distribution? On both debian 9 and debian unstable, they have:
>>> $ ls -l /dev/{,pts/}ptmx
>>> crw-rw-rw- 1 root tty 5, 2 mai 13 16:42 /dev/ptmx
>>> c--------- 1 root root 5, 2 mai 11 16:10 /dev/pts/ptmx
>>>
>>> and same on openSUSE Tumbleweed (20190402)
>>> On these two distros, the gcc tests have no problem when run as per the
>>> book.
--
http://lists.linuxfromscratch.org/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
