[LTP] [PATCH 1/1] aiocp.c: TCONF on O_DIRECT on tmpfs

[Richard Palethorpe]

Hello,

Petr Vorel <pvorel@suse.cz> writes:

>> Hi!
>> > > hm, it looks like it requires to skip tmpfs, as all tests TCONF:
>> > > aiocp.c:231: TINFO: Maximum AIO blocks: 65536
>> > > tst_device.c:541: TINFO: Use BTRFS specific strategy
>> > > tst_device.c:559: TBROK: BTRFS ioctl failed. Is . on a tmpfs?: ENOTTY (25)
>
>> > > I check if it'd work without
>> > > e1b1ae66b2 ("tst_find_backing_dev: Get dev name from /sys/dev/block/*/uevent")
>
>> > TMPDIR=/home/foo ./aiocp -b 1k -n 1 -f DIRECT # xfs
>> > tst_test.c:1558: TINFO: Timeout per run is 0h 30m 30s
>> > tst_test.c:1310: TINFO: xfs is supported by the test
>> > aiocp.c:231: TINFO: Maximum AIO blocks: 65536
>> > tst_device.c:585: TINFO: Use uevent strategy
>> > aiocp.c:267: TINFO: Fill srcfile.bin with random data
>> > aiocp.c:84: TBROK: write(3,0x7ffeccb9da00,60274) failed: EINVAL (22)
>
>> > TMPDIR=/var/tmp ./aiocp -b 1k -n 1 -f DIRECT # ext4
>> > tst_test.c:1558: TINFO: Timeout per run is 0h 30m 30s
>> > tst_test.c:1310: TINFO: ext2/ext3/ext4 is supported by the test
>> > aiocp.c:231: TINFO: Maximum AIO blocks: 65536
>> > tst_device.c:585: TINFO: Use uevent strategy
>> > aiocp.c:267: TINFO: Fill srcfile.bin with random data
>> > aiocp.c:84: TBROK: write(3,0x7ffda9e14120,50530) failed: EINVAL (22)
>
>> Ah, I suppose that this fails because we are wrongly passing the
>> srcflags to the open() that fills the source file with random data and
>> we end up passing O_DIRECT to the function that does not aling the
>> buffers and sizes at all.
>
>> Does this fix it?
>
> Hi Cyril,
>
> unfortunately the output is the same on all 3 tested filesystems.
> Any other hint what could fix it?
>

       EINVAL fd is attached to an object which is unsuitable for writing;  or
              the  file  was opened with the O_DIRECT flag, and either the ad-
              dress specified in buf, the value specified  in  count,  or  the
              file offset is not suitably aligned.

I suppose that writes should be aligned to the block size. Which
according to the lengths written they are not. There's also no reason to
open srcfd with O_DIRECT AFAICT?

We could also change fill_with_rand_data to use tst_rand_data. I'll post
a patch

>
> Kind regards,
> Petr
>
>> diff --git a/testcases/kernel/io/ltp-aiodio/aiocp.c b/testcases/kernel/io/ltp-aiodio/aiocp.c
>> index ee893ab11..a75da6a0c 100644
>> --- a/testcases/kernel/io/ltp-aiodio/aiocp.c
>> +++ b/testcases/kernel/io/ltp-aiodio/aiocp.c
>> @@ -263,7 +263,7 @@ static void setup(void)
>
>>         tst_res(TINFO, "Fill %s with random data", srcname);
>
>> -       srcfd = SAFE_OPEN(srcname, srcflags | O_RDWR | O_CREAT, 0666);
>> +       srcfd = SAFE_OPEN(srcname, O_WRONLY | O_CREAT, 0666);
>>         fill_with_rand_data(srcfd, filesize);
>>         SAFE_CLOSE(srcfd);
>>  }


-- 
Thank you,
Richard.