[LTP] [PATCH] mremap06: fallocate is not supported on nfsv3

[Petr Vorel]

Hi all,

> On 4/5/24 5:01 AM, Cyril Hrubis wrote:
> > Hi!
> > > The function fallocate() is not supported on nfsv3. Thus when we run the
> > > mremap06 test over a nfsv3 filesystem the test fails.

> > Can we rather than this just skip the test on nfsv3?

> > If we want to skip the test on nfs in generall we can just set
> > .skip_filesystems = {"nfs", NULL} in the tst_test structure.

> > I'm not sure if we can easily detect the nfs version. The test library
> > does that by checking the NFS_SUPER_MAGIC againts stat, but there is a
> > single SUPER_MAGIC for all nfs versions.

Yeah, include/uapi/linux/magic.h in kernel contains only single magic. BTW I
wonder if magic is stored to the disk or there is some other complication which
prevent to have more magic values which would allow to distinguish version.

I wonder if it makes sense to add support to detect NFS version (read vers
parameter in /proc/mounts). The only reason why to do that would be to make sure
NFS v4.2 did not lost fallocate support by bug/regression. If we don't care,
checking errno is better option (will reflect kernel changes without a need to
adapt LTP test).

> > So if we want to keep the test enabled for nfsv4 we can exit with TCONF
> > when the call fails only when we are on NFS with tst_fs_type() and check
> > that against TST_NFS_MAGIC.


> Hi Cyril,

> Wanted to mention that fallocate() is only supported over NFSv4.2 in
> particular. it's not supported over NFSv3, v4.0 or v4.1.
> I could even see on the fallocate ltp tests that it is handled the same
> manner.

Yeah, these tests have it:
testcases/kernel/mem/hugetlb/hugefallocate/hugefallocate01.c
testcases/kernel/mem/hugetlb/hugefallocate/hugefallocate02.c
testcases/kernel/syscalls/fallocate/fallocate01.c
testcases/kernel/syscalls/fallocate/fallocate02.c
testcases/kernel/syscalls/fallocate/fallocate03.c
testcases/kernel/syscalls/fallocate/fallocate04.c

I guess we should write SAFE_FALLOCATE(). @Samasth any change you would write
it (as a separate effort).

> Thanks,
> Samasth.

> > > Signed-off-by: Samasth Norway Ananda <samasth.norway.ananda@oracle.com>
> > > ---
> > >   testcases/kernel/syscalls/mremap/mremap06.c | 8 +++++++-
> > >   1 file changed, 7 insertions(+), 1 deletion(-)

> > > diff --git a/testcases/kernel/syscalls/mremap/mremap06.c b/testcases/kernel/syscalls/mremap/mremap06.c
> > > index 3bbaf441a..362b03e19 100644
> > > --- a/testcases/kernel/syscalls/mremap/mremap06.c
> > > +++ b/testcases/kernel/syscalls/mremap/mremap06.c
> > > @@ -104,8 +104,14 @@ static void setup(void)
> > >   	fd = SAFE_OPEN("testfile", O_CREAT | O_RDWR | O_TRUNC, 0600);
> > >   	ret = fallocate(fd, 0, 0, mmap_size);
> > > -	if (ret == -1)
> > > +	if (ret != 0) {
> > > +		if (errno == EOPNOTSUPP || errno == ENOSYS) {
> > > +			tst_brk(TCONF,
> > > +				"fallocate system call is not implemented");
> > > +		}
> > >   		tst_brk(TBROK, "fallocate() failed");
> > > +		return;

I suggest to merge this now, without return (it can be removed before merge).
Reviewed-by: Petr Vorel <pvorel@suse.cz>

Kind regards,
Petr