[RFC PATCH 2/2] landlock: selftests for bind and connect hooks

[Konstantin Meskhidze]

2/24/2022 12:55 PM, Mickaël Salaün пишет:
> 
> On 24/02/2022 04:18, Konstantin Meskhidze wrote:
>>
>>
>> 2/1/2022 9:31 PM, Mickaël Salaün пишет:
>>>
>>> On 24/01/2022 09:02, Konstantin Meskhidze wrote:
>>>> Support 4 tests for bind and connect networks actions:
>>>
>>> Good to see such tests!
>>>
>>>
>>>> 1. bind() a socket with no landlock restrictions.
>>>> 2. bind() sockets with landllock restrictions.
>>>
>>> You can leverage the FIXTURE_VARIANT helpers to factor out this kind 
>>> of tests (see ptrace_test.c).
>>>
>>>
>>>> 3. connect() a socket to listening one with no landlock restricitons.
>>>> 4. connect() sockets with landlock restrictions.
>>>
>>> Same here, you can factor out code. I guess you could create helpers 
>>> for client and server parts.
>>>
>>> We also need to test with IPv4, IPv6 and the AF_UNSPEC tricks.
>>>
>>> Please provide the kernel test coverage and explain why the uncovered 
>>> code cannot be covered: 
>>> https://www.kernel.org/doc/html/latest/dev-tools/gcov.html
>>
>>   Hi Mickaёl!
>>   Could you please provide the example of your test coverage build
>>   process? Cause as I undersatand there is no need to get coverage data
>>   for the entire kernel, just for landlock files.
> 
> You just need to follow the documentation:
> - start the VM with the kernel appropriately configured for coverage;
> - run all the Landlock tests;
> - gather the coverage and shutdown the VM;
> - use lcov and genhtml to create the web pages;
> - look at the coverage for security/landlock/
> 
    Thank you so much!

    One more questuoin - Is it possible to run Landlock tests in QEMU and
    and gather coverage info or I need to change kernel for the whole VM?
>>>
>>> You'll probably see that there are a multiple parts of the kernel 
>>> that are not covered. For instance, it is important to test different 
>>> combinations of layered network rules (see layout1/ruleset_overlap, 
>>> layer_rule_unions, non_overlapping_accesses, 
>>> interleaved_masked_accesses… in fs_test.c). Tests in fs_test.c are 
>>> more complex because handling file system rules is more complex, but 
>>> you can get some inspiration in it, especially the edge cases.
>>>
>>> We also need to test invalid user space supplied data (see 
>>> layout1/inval test in fs_test.c).
> .