Re: [10.0-amd64 host and 9.2-i386 jail] cpio: Can't update time for...

02.10.2013 23:12, Konstantin Belousov пишет:
> On Wed, Oct 02, 2013 at 10:18:44PM +0400, Boris Samorodov wrote:
>> 02.10.2013 21:47, Konstantin Belousov пишет:
>>> On Wed, Oct 02, 2013 at 09:27:29PM +0400, Boris Samorodov wrote:
>>>> (CCing to the maintainer)
>>>>
>>>> Hi All,
>>>>
>>>> I've come across a problem and can't diagnose it. Please give me
>>>> an advice how to proceed.
>>>>
>>>> I have a fresh CURRENT amd64 host with 9.2 (9.1 behaves the same)
>>>> i386 jail. The command "/usr/bin/find -d <input_dir> | /usr/bin/cpio
>>>> -dumpl <output_dir>" ends with error with diagnostic "cpio: Can't update
>>>> time for <a directory>".
>>>>
>>>> One can reproduce it by installing at 10-amd64 host poudriere, create
>>>> a 9-i386 jail and try to build devel/tmake port. Full log is here:
>>>> http://gw.wart.ru/bulk/91-i386-default/2013-10-02_14h52m08s/logs/errors/tmake-1.13.log
>>>>
>>>> I've managed to find out that the command...:
>>>> -----
>>>> # /usr/bin/find -d /wrkdirs/usr/ports/devel/tmake/work/tmake-1.13/lib
>>>> | /usr/bin/cpio -dumpl /destdir
>>>> -----
>>>>
>>>> ...fails with diagnostic: "cpio: Can't update time for
>>>> /destdir/wrkdirs/usr/ports/devel/tmake/work/tmake-1.13/lib".
>>>>
>>>> However the following command succeeds (mind the /* after lib):
>>>> -----
>>>> # /usr/bin/find -d /wrkdirs/usr/ports/devel/tmake/work/tmake-1.13/lib/*
>>>> | /usr/bin/cpio -dumpl /destdir
>>>> -----
>>>>
>>>> The directory itself seems quite natural:
>>>> -----
>>>> # ls -ldT /wrkdirs/usr/ports/devel/tmake/work/tmake-1.13/lib
>>>> drwxr-xr-x  53 root  wheel  3264 Jan 28 05:21:45 2004
>>>> /wrkdirs/usr/ports/devel/tmake/work/tmake-1.13/lib
>>>> -----
>>>>
>>>> There are no problems at 10-amd64 and 10-i386 jails.
>>>>
>>>> I'm out of ideas. Thanks for your help.
>>> ktrace the failing invocation ?
>>
>> The relevant part is here:
>> ftp://ftp.wart.ru/pub/misc/bsdcpio-error-cant-update-time-kdump.txt
> 
> So lutimes(2) fails with EINVAL, and in fact futimes(2) failed just before
> with the timeval array passed from the same address.  Most likely, EINVAL
> comes from the getutimes() check, which verifies that usec values are
> non-negative and less than 100000.
> 
> You probably have to debug this by looking at the timeval initialization
> and understanding where the value come from.

The bad value (for birthtim) came from lstat() syscall.

So far it seems to be a bug in time conversion. Pure hosts without
jails (both 10 and 9) are not affected. The simple testcase:
=====
host-amd64% sudo touch /jail/root/afile
host-amd64% ./test-utimes /jail/root/afile
---> /jail/root/afile
atim: 1380907943.417817881, mtim: 1380907943.417817881, ctim:
1380907943.417817881, birthtim:1380907943.417817881
...
jail-i386# ./test-utimes /root/afile
---> /root/afile
atim: 1380907943.417817881, mtim: 1380907943.417817881, ctim:
1380907943.417817881, birthtim:417817881.0
=====

Look at the birthtim results.

The testing program just do a stat/lstat (the resulta are the same)
and print the resulting values.

-- 
WBR, Boris Samorodov (bsam)
FreeBSD Committer, http://www.FreeBSD.org The Power To Serve