Re: [PATCH v3 2/5] selftests/mm: run_vmtests: remove sudo and conform to tap

[Ryan Roberts]

On 01/02/2024 12:24, Muhammad Usama Anjum wrote:
> On 2/1/24 5:04 PM, Ryan Roberts wrote:
>> On 25/01/2024 15:46, Muhammad Usama Anjum wrote:
>>> Remove sudo as some test running environments may not have sudo
>>> available. Instead skip the test if root privileges aren't available in
>>> the test.
>>>
>>> Signed-off-by: Muhammad Usama Anjum <usama.anjum@xxxxxxxxxxxxx>
>>> ---
>>> Changes since v1:
>>> - Added this patch in v2
>>>
>>> We are allocating 2*RLIMIT_MEMLOCK.rlim_max memory and mmap() isn't
>>> failing. This seems like true bug in the kernel. Even the root user
>>> shouldn't be able to allocate more memory than allowed MEMLOCKed memory.
>>> Any ideas?
>>> ---
>>>  tools/testing/selftests/mm/on-fault-limit.c | 36 ++++++++++-----------
>>>  tools/testing/selftests/mm/run_vmtests.sh   |  2 +-
>>>  2 files changed, 18 insertions(+), 20 deletions(-)
>>>
>>> diff --git a/tools/testing/selftests/mm/on-fault-limit.c b/tools/testing/selftests/mm/on-fault-limit.c
>>> index b5888d613f34e..0ea98ffab3589 100644
>>> --- a/tools/testing/selftests/mm/on-fault-limit.c
>>> +++ b/tools/testing/selftests/mm/on-fault-limit.c
>>> @@ -5,40 +5,38 @@
>>>  #include <string.h>
>>>  #include <sys/time.h>
>>>  #include <sys/resource.h>
>>> +#include "../kselftest.h"
>>>  
>>> -static int test_limit(void)
>>> +static void test_limit(void)
>>>  {
>>> -	int ret = 1;
>>>  	struct rlimit lims;
>>>  	void *map;
>>>  
>>> -	if (getrlimit(RLIMIT_MEMLOCK, &lims)) {
>>> -		perror("getrlimit");
>>> -		return ret;
>>> -	}
>>> +	if (getrlimit(RLIMIT_MEMLOCK, &lims))
>>> +		ksft_exit_fail_msg("getrlimit: %s\n", strerror(errno));
>>>  
>>> -	if (mlockall(MCL_ONFAULT | MCL_FUTURE)) {
>>> -		perror("mlockall");
>>> -		return ret;
>>> -	}
>>> +	if (mlockall(MCL_ONFAULT | MCL_FUTURE))
>>> +		ksft_exit_fail_msg("mlockall: %s\n", strerror(errno));
>>>  
>>>  	map = mmap(NULL, 2 * lims.rlim_max, PROT_READ | PROT_WRITE,
>>>  		   MAP_PRIVATE | MAP_ANONYMOUS | MAP_POPULATE, -1, 0);
>>> +
>>> +	ksft_test_result(map == MAP_FAILED, "Failed mmap\n");
>>> +
>>>  	if (map != MAP_FAILED)
>>> -		printf("mmap should have failed, but didn't\n");
>>> -	else {
>>> -		ret = 0;
>>>  		munmap(map, 2 * lims.rlim_max);
>>> -	}
>>> -
>>>  	munlockall();
>>> -	return ret;
>>>  }
>>>  
>>>  int main(int argc, char **argv)
>>>  {
>>> -	int ret = 0;
>>> +	ksft_print_header();
>>> +	ksft_set_plan(1);
>>> +
>>> +	if (getuid())
>>> +		ksft_test_result_skip("Require root privileges to run\n");
> I'd sent a patch to fix this behavior today. This test should run without
> root privileges.
> https://lore.kernel.org/all/20240201071307.592317-1-usama.anjum@xxxxxxxxxxxxx
> 
>>> +	else
>>> +		test_limit();
>>>  
>>> -	ret += test_limit();
>>> -	return ret;
>>> +	ksft_finished();
>>>  }
>>> diff --git a/tools/testing/selftests/mm/run_vmtests.sh b/tools/testing/selftests/mm/run_vmtests.sh
>>> index 246d53a5d7f28..e373d592dbf5c 100755
>>> --- a/tools/testing/selftests/mm/run_vmtests.sh
>>> +++ b/tools/testing/selftests/mm/run_vmtests.sh
>>> @@ -291,7 +291,7 @@ echo "$nr_hugepgs" > /proc/sys/vm/nr_hugepages
>>>  
>>>  CATEGORY="compaction" run_test ./compaction_test
>>>  
>>> -CATEGORY="mlock" run_test sudo -u nobody ./on-fault-limit
>>> +CATEGORY="mlock" run_test ./on-fault-limit
>>
>> I think changing this is going to give unintended results. run_vmtests.sh must
>> already be running as root. "sudo -u nobody" is deprivieging the test to run as
>> nobody. The rlimit is not enforced for root so this test must run unprivileged
>> to work. See man page for getrlimit():
>>
>>   Since Linux 2.6.9, no limits are placed on the amount of memory that a
>>   privileged process may lock, and this limit instead governs the amount of
>>   memory that an unprivileged process may lock
>>
>> So I think the correct fix is actually to install sudo on your CI.
> run_vmtests.sh is invoked without sudo with following:
> make -C tools/testing/selftests/mm run_tests

Unfortunately, I live in a world where my build machine isn't always the same as
the target machine, so I'm not too familiar with this method of invocation.

Regardless, the vast majority of the tests in run_vmtests.sh (as well as the
configuration code in the script itself) require root. So invoking
run_vmtests.sh as anything other than root is a BadIdea (TM). And when
run_vmtests.sh is running as root, then you need the "sudo -u nobody" to
deprivilege this particular test.

> 
> Installing sudo in rootfs wouldn't be trivial enough on the CI side.
> Alternatively, we can check if sudo is present before executing this test
> to avoid error that sudo isn't found.

Yeah, that's probably the easiest solution; just skip it if the requirements are
not met.

> 
>>
>>>  
>>>  CATEGORY="mmap" run_test ./map_populate
>>>  
>>
>>
>