[LTP] [RFC] fanotify: avoid TCONF on btrfs on TMPDIR

[Petr Vorel]

> On Tue, Apr 8, 2025 at 3:00 PM Petr Vorel <pvorel@suse.cz> wrote:
> >
> > Hi Amir, Jan,
> >
> > The subject was supposed to be "fanotify: avoid TCONF on btrfs on TMPDIR".
> >
> > Kind regards,
> > Petr
> >
> > > LTP tested on btrfs on TMPDIR is skipped with TCONF:
> > > # TMPDIR=/var/tmp ./fanotify23
> > > tst_tmpdir.c:316: TINFO: Using /var/tmp/LTP_fanzN3sOS as tmpdir (btrfs filesystem)
> > > tst_device.c:98: TINFO: Found free device 0 '/dev/loop0'
> > > tst_test.c:1194: TINFO: Formatting /dev/loop0 with ext2 opts='' extra opts=''
> > > mke2fs 1.47.0 (5-Feb-2023)
> > > tst_test.c:1206: TINFO: Mounting /dev/loop0 to /var/tmp/LTP_fanzN3sOS/fs_mnt fstyp=ext2 flags=0
> > > tst_test.c:1928: TINFO: LTP version: 20250130-190-g2c79e30c2c
> > > tst_test.c:1932: TINFO: Tested kernel: 6.14.0-3.g5e7754e-default #1 SMP PREEMPT_DYNAMIC Mon Mar 31 14:34:31 UTC 2025 (5e7754e) x86_64
> > > tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz'
> > > tst_kconfig.c:676: TINFO: CONFIG_FAULT_INJECTION kernel option detected which might slow the execution
> > > tst_test.c:1748: TINFO: Overall timeout per run is 0h 02m 00s
> > > fanotify23.c:233: TCONF: FAN_ATTRIB not supported on ext2 filesystem
> >
> > > The same is fanotify22.c
> > > fanotify22.c:296: TCONF: FAN_FS_ERROR not supported on ext4 filesystem
> >
> > > They work on tmpfs on TMPDIR:
> >
> > > # ./fanotify23
> > > tst_tmpdir.c:316: TINFO: Using /tmp/LTP_fanh6TUn4 as tmpdir (tmpfs filesystem)
> > > tst_device.c:98: TINFO: Found free device 0 '/dev/loop0'
> > > tst_test.c:1194: TINFO: Formatting /dev/loop0 with ext2 opts='' extra opts=''
> > > mke2fs 1.47.0 (5-Feb-2023)
> > > tst_test.c:1206: TINFO: Mounting /dev/loop0 to /tmp/LTP_fanh6TUn4/fs_mnt fstyp=ext2 flags=0
> > > tst_test.c:1928: TINFO: LTP version: 20250130-190-g2c79e30c2c
> > > tst_test.c:1932: TINFO: Tested kernel: 6.14.0-3.g5e7754e-default #1 SMP PREEMPT_DYNAMIC Mon Mar 31 14:34:31 UTC 2025 (5e7754e) x86_64
> > > tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz'
> > > tst_kconfig.c:676: TINFO: CONFIG_FAULT_INJECTION kernel option detected which might slow the execution
> > > tst_test.c:1748: TINFO: Overall timeout per run is 0h 02m 00s
> > > fanotify23.c:109: TPASS: FAN_MARK_ADD failed with EEXIST as expected when trying to downgrade to evictable mark
> > > fanotify23.c:72: TPASS: FAN_MARK_REMOVE failed with ENOENT as expected after empty mask
> > > fanotify23.c:155: TPASS: Got no events as expected
> > > fanotify23.c:72: TPASS: FAN_MARK_REMOVE failed with ENOENT as expected after drop_caches
> > > fanotify23.c:196: TPASS: got event: mask=4
> > > fanotify23.c:196: TPASS: got event: mask=4
> > > fanotify23.c:56: TINFO: Umounting /tmp/LTP_fanh6TUn4/fs_mnt
> > > fanotify23.c:57: TINFO: Mounting /dev/loop0 to /tmp/LTP_fanh6TUn4/fs_mnt fstyp=ext2 flags=0
> >
> > > Is there a way to avoid TCONF on btrfs?
> 
> I think that should do it:
> 
> --- a/testcases/kernel/syscalls/fanotify/fanotify22.c
> +++ b/testcases/kernel/syscalls/fanotify/fanotify22.c
> @@ -295,7 +295,7 @@ static void setup(void)
>  {
>         REQUIRE_FANOTIFY_EVENTS_SUPPORTED_ON_FS(FAN_CLASS_NOTIF|FAN_REPORT_FID,
>                                                 FAN_MARK_FILESYSTEM,
> -                                               FAN_FS_ERROR, ".");
> +                                               FAN_FS_ERROR, MOUNT_PATH);
>         pre_corrupt_fs();
> 
>         fd_notify = SAFE_FANOTIFY_INIT(FAN_CLASS_NOTIF|FAN_REPORT_FID,
> diff --git a/testcases/kernel/syscalls/fanotify/fanotify23.c
> b/testcases/kernel/syscalls/fanotify/fanotify23.c
> index 896a8a850..26c9e87fb 100644
> --- a/testcases/kernel/syscalls/fanotify/fanotify23.c
> +++ b/testcases/kernel/syscalls/fanotify/fanotify23.c
> @@ -229,10 +229,10 @@ static void setup(void)
>  {
>         SAFE_TOUCH(TEST_FILE, 0666, NULL);
> 
> -       REQUIRE_MARK_TYPE_SUPPORTED_ON_FS(FAN_MARK_EVICTABLE, ".");
> +       REQUIRE_MARK_TYPE_SUPPORTED_ON_FS(FAN_MARK_EVICTABLE, MOUNT_PATH);
>         REQUIRE_FANOTIFY_EVENTS_SUPPORTED_ON_FS(FAN_CLASS_NOTIF|FAN_REPORT_FID,
>                                                 FAN_MARK_FILESYSTEM,
> -                                               FAN_ATTRIB, ".");
> +                                               FAN_ATTRIB, MOUNT_PATH);
> 

> --

Thanks for a hint, indeed it works. I Cced you with a patch.
https://patchwork.ozlabs.org/project/ltp/patch/20250411132955.330226-1-pvorel@suse.cz/
https://lore.kernel.org/ltp/20250411132955.330226-1-pvorel@suse.cz/

> 
> I also see a bunch of:
> 
> $ git grep 'supported_on_fs.*\"\.\"'
> fanotify01.c:   if (fanotify_flags_supported_on_fs(FAN_REPORT_FID,
> FAN_MARK_MOUNT, FAN_OPEN, ".")) {
> fanotify05.c:   fd_error_unsupported =
> fanotify_init_flags_supported_on_fs(FAN_REPORT_FD_ERROR, ".");
> fanotify11.c:   fan_report_tid_unsupported =
> fanotify_init_flags_supported_on_fs(FAN_REPORT_TID, ".");
> fanotify21.c:           fd_error_unsupported =
> fanotify_init_flags_supported_on_fs(FAN_REPORT_FD_ERROR, ".");
> 
> The first one is intentional.
> The two FAN_REPORT_FD_ERROR ones could use MOUNT_PATH,
> but I don't think that is critical.

+1, therefore I kept it the same.

Kind regards,
Petr

> There is nothing about those flags that is fs specific, so it is a
> pure kernel feature test.
> 
> The problem with fanotify{22,23} was with the flag FAN_MARK_FILESYSTEM that
> is not supported on a btrfs subvolume.
> 
> Thanks,
> Amir.