Possible issue around utime() and symlinks on at least MIPS64

Possible issue around utime() and symlinks on at least MIPS64

[Adam Hill]

[-- Attachment #1: Type: text/plain, Size: 951 bytes --]

Hi,

	Not carried out ANY investigation around this... unfortunately I don't have any time 
atm. Could also have been mentioned already or actually be a bug in tar itself etc. but I 
couldn't find anything with my quick search. Just thought I'd mention it in case anyone does 
feel inclined to investigate.

	I compiled OpenWrt using pretty much default build settings for an octeon chip... it 
uses musl-1.1.16. The only real change I made was to use GNU tar rather than busybox. The 
resulting tar binary can't fix the modification/access times on any extracted symbolic links, 
resulting in for example:

tar: usr/share/zoneinfo/Africa: Cannot utime: Invalid argument

	I didn't have strace available so can't even be sure tar's using utime()... just going 
by the error message alone.

	I recompiled OpenWrt changing only the C library ( to GCC ) and the resulting tar 
binary worked fine... so I can only *assume* it's something to do with musl.

[-- Attachment #2: Type: text/html, Size: 2870 bytes --]

Re: Possible issue around utime() and symlinks on at least MIPS64

[Rich Felker]

On Thu, Jul 06, 2017 at 02:50:16PM +0100, Adam Hill wrote:
> Hi,
> 
> 	Not carried out ANY investigation around this... unfortunately I don't have any time 
> atm. Could also have been mentioned already or actually be a bug in tar itself etc. but I 
> couldn't find anything with my quick search. Just thought I'd mention it in case anyone does 
> feel inclined to investigate.
> 
> 	I compiled OpenWrt using pretty much default build settings for an octeon chip... it 
> uses musl-1.1.16. The only real change I made was to use GNU tar rather than busybox. The 
> resulting tar binary can't fix the modification/access times on any extracted symbolic links, 
> resulting in for example:
> 
> tar: usr/share/zoneinfo/Africa: Cannot utime: Invalid argument
> 
> 	I didn't have strace available so can't even be sure tar's using utime()... just going 
> by the error message alone.
> 
> 	I recompiled OpenWrt changing only the C library ( to GCC ) and the resulting tar 
> binary worked fine... so I can only *assume* it's something to do with musl.

Did anyone ever get a chance to look further into this? It's been on
my todo for a long time, but without sufficient information to
reproduce I don't know where to begin. I spent about 20 minutes
looking at the musl and kernel code now and it does not seem likely
that this is mips64-specific.

Rich

Re: Possible issue around utime() and symlinks on at least MIPS64

[A. Wilcox]

[-- Attachment #1.1: Type: text/plain, Size: 1894 bytes --]

On 04/18/18 15:00, Rich Felker wrote:
> On Thu, Jul 06, 2017 at 02:50:16PM +0100, Adam Hill wrote:
>> Hi,
>> 
>> Not carried out ANY investigation around this... unfortunately I
>> don't have any time atm. Could also have been mentioned already or
>> actually be a bug in tar itself etc. but I couldn't find anything
>> with my quick search. Just thought I'd mention it in case anyone
>> does feel inclined to investigate.
>> 
>> I compiled OpenWrt using pretty much default build settings for an
>> octeon chip... it uses musl-1.1.16. The only real change I made was
>> to use GNU tar rather than busybox. The resulting tar binary can't
>> fix the modification/access times on any extracted symbolic links,
>>  resulting in for example:
>> 
>> tar: usr/share/zoneinfo/Africa: Cannot utime: Invalid argument
>> 
>> I didn't have strace available so can't even be sure tar's using
>> utime()... just going by the error message alone.
>> 
>> I recompiled OpenWrt changing only the C library ( to GCC ) and the
>> resulting tar binary worked fine... so I can only *assume* it's
>> something to do with musl.
> 
> Did anyone ever get a chance to look further into this? It's been on 
> my todo for a long time, but without sufficient information to 
> reproduce I don't know where to begin. I spent about 20 minutes 
> looking at the musl and kernel code now and it does not seem likely 
> that this is mips64-specific.
> 
> Rich
> 

This really sounds like broken cross-compiling.  When we cross-compile
cp, it's unable to handle xattrs until it's re-compiled on native
hardware.  I'd suggest poking around at autoconf for GNU tar and seeing
what it is 'guessing' the environment is like, and perhaps overriding
some with ac_*={yes,no} on the configure line.

Best,
--arw

-- 
A. Wilcox (awilfox)
Project Lead, Adélie Linux
http://adelielinux.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: Possible issue around utime() and symlinks on at least MIPS64

[Rich Felker]

On Wed, Apr 18, 2018 at 04:51:31PM -0500, A. Wilcox wrote:
> On 04/18/18 15:00, Rich Felker wrote:
> > On Thu, Jul 06, 2017 at 02:50:16PM +0100, Adam Hill wrote:
> >> Hi,
> >> 
> >> Not carried out ANY investigation around this... unfortunately I
> >> don't have any time atm. Could also have been mentioned already or
> >> actually be a bug in tar itself etc. but I couldn't find anything
> >> with my quick search. Just thought I'd mention it in case anyone
> >> does feel inclined to investigate.
> >> 
> >> I compiled OpenWrt using pretty much default build settings for an
> >> octeon chip... it uses musl-1.1.16. The only real change I made was
> >> to use GNU tar rather than busybox. The resulting tar binary can't
> >> fix the modification/access times on any extracted symbolic links,
> >>  resulting in for example:
> >> 
> >> tar: usr/share/zoneinfo/Africa: Cannot utime: Invalid argument
> >> 
> >> I didn't have strace available so can't even be sure tar's using
> >> utime()... just going by the error message alone.
> >> 
> >> I recompiled OpenWrt changing only the C library ( to GCC ) and the
> >> resulting tar binary worked fine... so I can only *assume* it's
> >> something to do with musl.
> > 
> > Did anyone ever get a chance to look further into this? It's been on 
> > my todo for a long time, but without sufficient information to 
> > reproduce I don't know where to begin. I spent about 20 minutes 
> > looking at the musl and kernel code now and it does not seem likely 
> > that this is mips64-specific.
> > 
> > Rich
> > 
> 
> This really sounds like broken cross-compiling.  When we cross-compile
> cp, it's unable to handle xattrs until it's re-compiled on native
> hardware.  I'd suggest poking around at autoconf for GNU tar and seeing
> what it is 'guessing' the environment is like, and perhaps overriding
> some with ac_*={yes,no} on the configure line.

That's a good thought and would probably explain why we haven't hit a
similar issue elsewhere. It probably also indicates a bug in
gnulib/tar rather than musl.

Rich