[linux-lvm] [lvm-devel] MUSL fun

[Zdenek Kabelac]

Bernd,
[cc buildroot]
Like you I'd been investigating this issue from the buildroot end.
Since October I've had no spare time to look at buildroot fixes.
I was hoping for an explanation about why LVM implemented the above
behaviour, but I never got one, and I didn't fancy working on a patch
that would never get accepted when other projects were more open ..
I think the best approach is probably just to add the freopen() patch
as Romain suggested and keep the patch available in buildroot in the
hope that we can build support for LVM to accept it.
Brendan

if (!freopen(NULL, "r", stdin))
goto out;
.. but I suspect there is a good reason why this isn't done. Can anyone
shed any light on this ? If not I will submit a patch.

Hi Brendan,
the buildroot project also supports musl, the build of lvm2 fails due to
the problem you described. Did you find a working solution for the
problem? A patch I submitted to the buildroot project, based on patches I
found at the Alpinelinux project, raised some questions I am unable to
answer: http://patchwork.ozlabs.org/patch/573519/
Regards, Bernd

Let's start from the beginning -
--
commands/toolcontext.c: In function 'create_toolcontext':
commands/toolcontext.c:1796:10: error: assignment of read-only variable 'stdin'
stdin = new_stream;
--
So now - from where comes the idea of 'stdin' being read-only variable ?
Looking at some POSIX spec e.g.:
http://pubs.opengroup.org/onlinepubs/009695399/functions/stdin.html

there is no sign of having this read-only var.

So what kind of OS this actually is ?


There is a reason from lvm2 implementation here - freopen() is not giving
application full control over internal buffer allocation and we need to be
sure we lock-in memory for critical section - and some glibc versions are
allocating buffers here via 'mmap' call.

That said - there could be accepted a patch checking in configure for
'read-only' stdin - and using then #ifdef compilation that would
replace use of internal lvm2 reopen code with libc function.

But the currently posted patch may possibly cause some other regressions.

Regards

Zdenek

[Zdenek Kabelac]

Post by Zdenek Kabelac
There is a reason from lvm2 implementation here - freopen() is not giving
application full control over internal buffer allocation and we need to be
sure we lock-in memory for critical section - and some glibc versions are
allocating buffers here via 'mmap' call.
That said - there could be accepted a patch checking in configure for
'read-only' stdin - and using then #ifdef compilation that would
replace use of internal lvm2 reopen code with libc function.

I had a suspicion you would have a good reason.
On first glance your proposal sounds like a good compromise and one
that would be worth exploring.

As an alternative idea, what about changing the codebase not to use
stdin/stdout/stderr, using it's own file descriptor names and calling
dup2() as appropriate to set them up on startup ?
Just flying the kite - I haven't looked to see how painful this might be.

Hi

Well don't know anyone who would play with this approach which is several
orders in magnitude more difficult then the plain and simple #ifdef.

I also assume we may probably even overwrite value of stdin via tricky pointer
magic so compiler will not be able to spot mods of stdin in buildtime.

Regards

Zdenek