Message-ID: <807bb98f-4aa6-4e15-8dd4-5d27914f9910@gmail.com>
Date: Tue, 18 Mar 2025 10:27:41 +0530
From: "Nirjhar Roy (IBM)" <nirjhar.roy.lists@...il.com>
To: "Darrick J. Wong" <djwong@...nel.org>
Cc: fstests@...r.kernel.org, linux-ext4@...r.kernel.org,
 linux-xfs@...r.kernel.org, ritesh.list@...il.com, ojaswin@...ux.ibm.com,
 zlang@...nel.org, david@...morbit.com
Subject: Re: [PATCH v1 2/2] check,common/{preamble,rc},soak: Decoupling
 init_rc() call from sourcing common/rc


On 3/12/25 11:16, Nirjhar Roy (IBM) wrote:
>
> On 3/7/25 23:10, Darrick J. Wong wrote:
>> On Fri, Mar 07, 2025 at 11:21:15AM +0530, Nirjhar Roy (IBM) wrote:
>>> On 3/6/25 23:16, Darrick J. Wong wrote:
>>>> On Thu, Mar 06, 2025 at 08:17:41AM +0000, Nirjhar Roy (IBM) wrote:
>>>>> Silently executing scripts during sourcing common/rc doesn't look 
>>>>> good
>>>>> and also causes unnecessary script execution. Decouple init_rc() call
>>>>> and call init_rc() explicitly where required.
>>>>>
>>>>> Signed-off-by: Nirjhar Roy (IBM) <nirjhar.roy.lists@...il.com>
>>>>> ---
>>>>>    check           | 10 ++--------
>>>>>    common/preamble |  1 +
>>>>>    common/rc       |  2 --
>>>>>    soak            |  1 +
>>>>>    4 files changed, 4 insertions(+), 10 deletions(-)
>>>>>
>>>>> diff --git a/check b/check
>>>>> index ea92b0d6..d30af1ba 100755
>>>>> --- a/check
>>>>> +++ b/check
>>>>> @@ -840,16 +840,8 @@ function run_section()
>>>>>            _prepare_test_list
>>>>>        elif [ "$OLD_TEST_FS_MOUNT_OPTS" != "$TEST_FS_MOUNT_OPTS" 
>>>>> ]; then
>>>>>            _test_unmount 2> /dev/null
>>>>> -        if ! _test_mount
>>>>> -        then
>>>>> -            echo "check: failed to mount $TEST_DEV on $TEST_DIR"
>>>>> -            status=1
>>>>> -            exit
>>>>> -        fi
>>>> Unrelated change?  I was expecting a mechanical ". ./common/rc" =>
>>>> ". ./common/rc ; init_rc" change in this patch.
>>> This patch adds an init_rc() call to _begin_fstests() in 
>>> common/preamble and
>>> hence the above _test_mount() will be executed during that call. So 
>>> this
>>> _test_mount isn't necessary here, right? _test_mount() will be 
>>> executed (as
>>> a part of init_rc() call) before every test run. Please let me know 
>>> if my
>>> understanding isn't correct.
>> It's true that in terms of getting the test filesystem mounted, the
>> _test_mount here and in init_rc are redundant.  But look at what happens
>> on error here -- we print "check: failed to mount..." to signal that the
>> new section's TEST_FS_MOUNT_OPTS are not valid, and exit the ./check
>> process.
>>
>> By deferring the mount to the init_rc in _preamble, that means that
>> we'll run the whole section with bad mount options, most likely
>> resulting in every test spewing "common/rc: could not mount..." and
>> appearing to fail.
> Aah, right. The exit should be at the check level if some parameter is 
> not correct in a section. I will make the change in v2.
>>
>> I think.  I'm not sure what "status=1; exit" does as compared to
>> "exit 1"; AFAICT the former actually results in an exit code of 0
>> because the (otherwise pointless) assignment succeeds.
>
> I think "status=0; exit" has a reason. If we see the following trap 
> handler registration in the check script:
>
> if $OPTIONS_HAVE_SECTIONS; then
>     trap "_kill_seq; _summary; exit \$status" 0 1 2 3 15
> else
>     trap "_kill_seq; _wrapup; exit \$status" 0 1 2 3 15
> fi
>
> So, "exit 1" will exit the check script without setting the correct 
> return value. I ran with the following local.config file:
>
> [xfs_4k_valid]
> FSTYP=xfs
> TEST_DEV=/dev/loop0
> TEST_DIR=/mnt1/test
> SCRATCH_DEV=/dev/loop1
> SCRATCH_MNT=/mnt1/scratch
>
> [xfs_4k_invalid]
> FSTYP=xfs
> TEST_DEV=/dev/loop0
> TEST_DIR=/mnt1/invalid_dir
> SCRATCH_DEV=/dev/loop1
> SCRATCH_MNT=/mnt1/scratch
>
> This caused the init_rc() to catch the case of invalid _test_mount 
> options. Although the check script correctly failed during the 
> execution of the "xfs_4k_invalid" section, the return value was 0, i.e 
> "echo $?" returned 0. This is because init_rc exits with "exit 1" 
> without correctly setting the value of "status".
>
> However, when I executed with the following local.config file:
>
> [xfs_4k_valid]
> FSTYP=xfs
> TEST_DEV=/dev/loop0
> TEST_DIR=/mnt1/test
> SCRATCH_DEV=/dev/loop1
> SCRATCH_MNT=/mnt1/scratch
>
> [xfs_4k_invalid]
> FSTYP=xfs
> TEST_DEV=/dev/loop0
> TEST_DIR=/mnt1/invalid_dir
> SCRATCH_DEV=/dev/loop1
> SCRATCH_MNT=/mnt1/scratch
> TEST_FS_MOUNT_OPTS="-o invalidss"
>
> This caused the "elif [ "$OLD_TEST_FS_MOUNT_OPTS" != 
> "$TEST_FS_MOUNT_OPTS" ]; then" to be executed. Now, when I checked the 
> value of "echo $?", it showed 1. IMO, this is the correct behavior, 
> and we should always use "status=<value>; exit" and NOT "exit 1" 
> directly. Even if 1 section fails,   "./check <test-list>" command 
> should return a non-zero value. Can you please let me know if my 
> understanding is correct? If yes, maybe we can have a function like
>
> _set_status_and_exit()
> {
>
>     status="$1"
>     exit
> }
>
> and replace all the "status <value>; exit" and "exit <value>" with 
> "_set_status_and_exit <value>"

Hi Darrick, any feedback on the above?

--NR

>
> --NR
>
>
>>
>> Granted, the init_rc that you remove below would also catch that case
>> and exit ./check
> Yes. init_rc can catch that case with an additional difference that it 
> will attempt another mount "retrying test device mount with external set"
>>
>>>>>        fi
>>>>> -    init_rc
>>>> Why remove init_rc here?
>>> Same reason as above.
>> But that's an additional change in behavior.  If there's no reason for
>> calling init_rc() from run_section() then that should be a separate
>> patch with a separate justification.
>
> Since the check for _test_mount should be at the check script level 
> and not at the _begin_fstest(), maybe we should
>
> 1. Keep the init_rc call here
>
> 2. Remove the _test_mount above (the one with "elif [ 
> "$OLD_TEST_FS_MOUNT_OPTS" != "$TEST_FS_MOUNT_OPTS" ]; then" ) and have 
> a separate patch for it with proper justification.
>
> 3. NOT have any init_rc call in _begin_fstest(), since the _test_mount 
> related checks would already been done by the time _begin_fstests() 
> gets executed.
>
> The above changes will also not change any existing behavior. Can you 
> please let me know your thoughts and I can send a V2 accordingly?
>
> --NR
>
>>
>> --D
>>
>>>>> -
>>>>>        seq="check.$$"
>>>>>        check="$RESULT_BASE/check"
>>>>> @@ -870,6 +862,8 @@ function run_section()
>>>>>        needwrap=true
>>>>>        if [ ! -z "$SCRATCH_DEV" ]; then
>>>>> +        _check_mounted_on SCRATCH_DEV $SCRATCH_DEV SCRATCH_MNT 
>>>>> $SCRATCH_MNT
>>>>> +        [ $? -le 1 ] || exit 1
>>>>>          _scratch_unmount 2> /dev/null
>>>>>          # call the overridden mkfs - make sure the FS is built
>>>>>          # the same as we'll create it later.
>>>>> diff --git a/common/preamble b/common/preamble
>>>>> index 0c9ee2e0..c92e55bb 100644
>>>>> --- a/common/preamble
>>>>> +++ b/common/preamble
>>>>> @@ -50,6 +50,7 @@ _begin_fstest()
>>>>>        _register_cleanup _cleanup
>>>>>        . ./common/rc
>>>>> +    init_rc
>>>>>        # remove previous $seqres.full before test
>>>>>        rm -f $seqres.full $seqres.hints
>>>>> diff --git a/common/rc b/common/rc
>>>>> index d2de8588..f153ad81 100644
>>>>> --- a/common/rc
>>>>> +++ b/common/rc
>>>>> @@ -5754,8 +5754,6 @@ _require_program() {
>>>>>        _have_program "$1" || _notrun "$tag required"
>>>>>    }
>>>>> -init_rc
>>>>> -
>>>>> ################################################################################
>>>>>    # make sure this script returns success
>>>>>    /bin/true
>>>>> diff --git a/soak b/soak
>>>>> index d5c4229a..5734d854 100755
>>>>> --- a/soak
>>>>> +++ b/soak
>>>>> @@ -5,6 +5,7 @@
>>>>>    # get standard environment, filters and checks
>>>>>    . ./common/rc
>>>>> +# ToDo: Do we need an init_rc() here? How is soak used?
>>>> I have no idea what soak does and have never used it, but I think for
>>>> continuity's sake you should call init_rc here.
>>> Okay. I think Dave has suggested removing this file[1]. This doesn't 
>>> seem to
>>> used anymore.
>>>
>>> [1] https://lore.kernel.org/all/Z8oT_tBYG-a79CjA@dread.disaster.area/
>>>
>>> --NR
>>>
>>>> --D
>>>>
>>>>>    . ./common/filter
>>>>>    tmp=/tmp/$$
>>>>> -- 
>>>>> 2.34.1
>>>>>
>>>>>
>>> -- 
>>> Nirjhar Roy
>>> Linux Kernel Developer
>>> IBM, Bangalore
>>>
>>>
-- 
Nirjhar Roy
Linux Kernel Developer
IBM, Bangalore

Powered by blists - more mailing lists