On Tue, Feb 14, 2017 at 6:40 AM, Xiong Zhou <xzhou@xxxxxxxxxx> wrote:
> On Mon, Feb 13, 2017 at 07:37:45AM +0200, Amir Goldstein wrote:
>> On Mon, Feb 13, 2017 at 6:19 AM, Xiong Zhou <xzhou@xxxxxxxxxx> wrote:
>> > On Sun, Feb 12, 2017 at 10:43:35PM +0200, Amir Goldstein wrote:
>> >> Hi Eryu and all,
>> >>
>> >> The reason I started this work was to help catch overlayfs bugs
>> >> related to leaking objects in underlying (base) fs.
>> >
>> > So firstly, what's wrong with the existing way exactly ?
>> >
>>
>> One of the reasons that xfstests cycle mounts scratch partition
>> is that some bugs, such as object leaking bugs, are only discovered
>> when trying to unmount the file system.
>>
>> With overlayfs tests the cycling of overlay mount can detect bugs
>> related to overlayfs objects leaking, but not to base fs objects leaking.
>> OTOH, some bugs may be triggered in base fs when used under
>> overlayfs, but not when tests are run directly over the fs.
>>
>> The solution is to cycle mount both overlay mount and the base fs
>> underneath it to detect all those bugs.
>
> Good job.
>
>>
>> I started off (in V1) with a more complex way to manually configure
>> OVL_BASE_TEST_DEV, OVL_BASE_TEST_DIR, etc, but that
>> became very cumbersome.
>>
>> Eryu pointed out the over complication of this configuration, so
>> I decided to re-think the "old way", which I always thought was
>> a bit hackish.
>>
>> The "new way" enables running overlay tests over a range of
>> base fs configurations, which was not possible before without
>> using external scripts to wrap xfstests.
>>
>> But above all, the "new way" represents a different perspective
>> of overlayfs testing - testing overlayfs independently of the
>> underlying fs is doing half the job.
>>
>> You may say that overlayfs is closer to "MOUNT_OPTIONS"
>> then it is to "FSTYP". The truth is somewhere in the middle,
>> but the fact is that running a single "overlay" test does not cut it.
>> To tests overlay thoroughly, one would need to test overlay over
>> xfs, ext4, btrfs and to test xfs thoroughly, one would need to test
>> xfs, xfs+reflink, xfs+overlay, xfs+reflink+overlay, etc.
>>
>> The new way provides an easy way to configure those tests,
>> using semantics that people are used to when testing multiple
>> flavors of configurations.
>
> I'd like more comments or document for these NEW configurations, other
> than one or two lines in Examples.
>
> In other words, what kind of configuration your patches bringing
> support to ?
>
There are 2 parts for this answer:
1. Single section config file
====================
The answer here is there is no new configuration.
All you need to do is forget everything you knew about old overlay test config.
Take an existing test configuration you have and already use:
TEST_DEV=/dev/sda5
TEST_DIR=/mnt/test
SCRATCH_DEV=/dev/sda6
SCRATCH_MNT=/mnt/scratch
FSTYP=xfs
and just run './check -overlay' whenever you want to test overlayfs
and './check' whenever you want to test native fs.
I could add more and more examples to README, but it's not like
README tells you much about how to use other configurations.
IMO, these 3 lines sums it up well:
- for overlay tests: ./check -overlay [test(s)]
The TEST and SCRATCH partitions should be pre-formatted
with another base fs, where the overlay dirs will be created
2. Multi section config file
===================
Honestly, I never intended to make it work and there are many combinations
of sections that do not work today and some that will not work with overlay
(see email from Ted about patch 9/9)
I just realized that it may be useful to some and added the example howto
add overlay sections post non-overlay section, but I do not intend to fix
problems with this sort of setup, so I might as well remove the example
and leave this an undocumented feature for people who understand
how multi section works.
--
To unsubscribe from this list: send the line "unsubscribe linux-unionfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
