Re: [PATCH] squash4: fix handling of fragments and sparse files

[Andrei Borzenkov]

23.02.2017 16:52, Vladimir 'phcoder' Serbinenko пишет:
> On Sat, Feb 18, 2017, 10:17 Andrei Borzenkov <address@hidden> wrote:
> 
>> 1. Do not assume block list and fragment are mutually exclusive. Squash
>> can pack file tail as fragment (unless -no-fragments is specified); so
>> check read offset and read either from block list or from fragments as
>> appropriate.
>>
>> 2. Support sparse files with zero blocks.
>>
>> 3. Fix fragment read - frag.offset is absolute fragment position,
>> not offset relative to ino.chunk.
>>
> Go ahead.
> 

Done with cosmetic changes (renamed one variable to better match
surrounding code and removed now superfluous assignment).

>>
>> Reported and tested by Carlo Caione <address@hidden>
>>
>> ---
>>
>> @Vladimir: we need regression tests for both of these cases. We could real
>> small
>> files only by accident - block list was zero, so it appeared to work.
>>
> How do you create those files reliably? Feel free to add any files to
> grub-fs-tester. Feel free to commit without tests, we can add tests later.
> 

I would say, we need to generally create less "nice" test files

1. Generate test files that are not exact multiple of filesystem block.
This would cover squash tail packing and may uncover similar issues in
other drivers as well.

2. Generate sparse files by seeking beyond end of file. We already had
similar problem with indirect ext* blocks that was not caught by tests.
We should produce holes both for direct and indirect blocks (we may pick
common offset or make it fs-specific if necessary).

3. We need to produce really small files also to test inlining. Again we
can pick common size or make it fs-specific.

Assuming we generate files as described, for squash4 just call it with
-always-use-fragments to force tail packing. -nopad may be interesting
as well to stress our driver even more.