ftruncate04 broken on kernels without mandatory locking

Upstream kernel commit

9e8925b67a809bb27ce4b7d352d67f25cf1d7fc5
locks: Allow disabling mandatory locking at compile time

added a config option to remove support for mandatory locking (mount -o mand, MS_MANDLOCK), which went into v4.5.

Some distributions (like Fedora) already disable it, causing ftruncate04 to fail:

ftruncate04    0  TINFO  :  TMPDIR does not support mandatory locks
ftruncate04    0  TINFO  :  Found free device '/dev/loop0'
ftruncate04    0  TINFO  :  Formatting /dev/loop0 with ext2 opts=''

extra opts=''
mke2fs 1.42.13 (17-May-2015)
ftruncate04 1 TBROK : safe_macros.c:728: ftruncate04.c:247:
mount(/dev/loop0, dir/, ext2, 64, (nil)) failed: errno=EPERM(1):
Operation not permitted
ftruncate04 2 TBROK : safe_macros.c:728: Remaining cases broken

and indeed

$ mount /dev/loop0 /mnt/
$ umount /mnt
$ mount /dev/loop0 /mnt/ -o mand
mount: permission denied

The question is how to best fix the testcase - do the mount unsafely and assume EPERM should be TCONF? Or somehow check kernel config? Maybe do the mount first without MS_MANDLOCK (to rule out other perm issues) and then remount with MS_MANDLOCK, checking EPERM?

Any other ideas?

Thanks,
Jiri

Hi!

Too bad that it returns EPERM rather than ENOSYS.

Parsing kernel config portably between major distributions is very troublesome if not impossible so I would like to avoid this path.

Checking for EPERM while remounting with MS_MANDLOCK sounds good to me.

Cyril Hrubis
[email protected]