[CRIU] [RFC PATCH 01/21] zdtm: zdtm.py: elaborate file copying logic for ns tests

[Pavel Emelyanov]

On 03/09/2016 02:33 PM, Ivan Shapovalov wrote:
> On 2016-03-09 at 13:18 +0300, Pavel Emelyanov wrote:
>> On 02/26/2016 01:27 AM, Ivan Shapovalov wrote:
>>>
>>> On 2016-02-25 at 12:01 -0800, Andrew Vagin wrote:
>>>>
>>>> On Fri, Feb 19, 2016 at 05:50:24PM +0300, Ivan Shapovalov wrote:
>>>>>
>>>>>
>>>>> From: Ivan Shapovalov <intelfx100 at gmail.com>
>>>>>
>>>>> When copying files (tests' deps and supplementary binaries) to
>>>>> the
>>>>> "chroot",
>>>>> do not blindly copy them using provided pathes (like
>>>>> "/sbin/iptables").
>>>>> Instead, walk the path manually and replicate all encountered
>>>>> symlinks in
>>>>> the "chroot".
>>>>>
>>>>> This helps to support setups like Arch, where /{,s}bin are
>>>>> links to
>>>>> /usr/bin,
>>>>> and only /usr/bin is in $PATH.
>>>> Why is it not enough to call os.path.realpath?
>>> Consider the following example:
>>>
>>> - /bin, /sbin are symlinks to /usr/bin
>>> - /usr/bin/iptables is a symlink to xtables-multi
>>> - $PATH is /usr/bin (just this one directory)
>>> - test says /sbin/iptables in deps
>> And we'll copy /sbin/iptables to /sbin/iptables and will get
>> the binary in test fake root. If the problem is only in $PATH,
>> maybe it's worth fixing _it_?
> 
> Can't something break if an executable is moved to a different prefix?

In theory it can, but so far we haven't met anything like this. Moved
binary is almost always fixable with proper PATH variable :)

> (I agree that my patch is way too overcomplicated. If simply force-
> setting $PATH to something like /sbin:/bin:/usr/sbin:/usr/bin is OK,
> then I'll happily drop this patch and do that instead.)
> 
> --
> Ivan Shapovalov / intelfx /
> 
>>
>>>
>>> Just realpath will copy /usr/bin/xtables-multi and there will be no
>>> iptables at all.
>>>
>>> realpath plus naïve symlink will copy /usr/bin/xtables-multi and
>>> make
>>> /sbin/iptables a link to it, and iptables will be missing from
>>> $PATH.
>>>
>>>
>>>
>>> _______________________________________________
>>> CRIU mailing list
>>> CRIU at openvz.org
>>> https://lists.openvz.org/mailman/listinfo/criu