btrfs: indicate iversion option in show_options
PATCH 1/1 btrfs: indicate iversion option in show_options
Eric reported a problem where if you did
mount -o remount /some/btrfs/fs
you would lose SB_I_VERSION on the mountpoint. After a very convoluted search I discovered this is because the remount infrastructure doesn't just say "change these things specifically", but it actually depends on userspace to tell it fucking everything that needs to be set on the mountpoint. This led to the fucking horrifying discovery that util-linux actually has to parse /proc/mounts to figure out what the fuck is set on the mount point in order to preserve any of the options it's not actually fucking with, so in this case iversion. If we don't indicate iversion is set, then we get iversion cleared on the mount, because util-linux doesn't pass in MS_I_VERSION as it's mount flags.
So work around this fucking insanity by spitting out iversion in /proc/mounts so we get the correct flags passed to us in remount.
--- fs/btrfs/super.c | 4 ++++ 1 file changed, 4 insertions(+)
fs/btrfs/super.c +4 -0
@@ -1427,6 +1427,10 @@ static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)seq_puts (seq,",discard=async" );if (!(info->sb->s_flags &SB_POSIXACL ))seq_puts (seq,",noacl" ); +if (info->sb->s_flags &SB_I_VERSION ) +seq_puts (seq,",iversion" ); +else +seq_puts (seq,",noiversion" );if (btrfs_test_opt (info,SPACE_CACHE ))seq_puts (seq,",space_cache" );else if (btrfs_test_opt (info,FREE_SPACE_TREE ))
Signature · 1 line
2.24.1
Quoted 15 lines · Josef Bacik
Eric reported a problem where if you did
mount -o remount /some/btrfs/fs
you would lose
SB_I_VERSIONon the mountpoint. After a very convoluted search I discovered this is because the remount infrastructure doesn't just say "change these things specifically", but it actually depends on userspace to tell it fucking everything that needs to be set on the mountpoint. This led to the fucking horrifying discovery that util-linux actually has to parse/proc/mountsto figure out what the fuck is set on the mount point in order to preserve any of the options it's not actually fucking with, so in this case iversion. If we don't indicate iversion is set, then we get iversion cleared on the mount, because util-linux doesn't pass inMS_I_VERSIONas it's mount flags.
Quoted 2 lines
So work around this fucking insanity by spitting out iversion in
/proc/mountsso we get the correct flags passed to us in remount.
Hmmm:
# mount -o loop,noiversion btrfsfile mnt # grep btrfs /proc/mounts /dev/loop0 /tmp/mnt btrfs rw,seclabel,relatime,iversion,space_cache,subvolid=5,subvol=/ 0 0 #
Quoted 22 lines
Reported-by Eric Sandeen <[email protected]>Signed-off-by Josef Bacik <[email protected]>--- fs/btrfs/super.c | 4 ++++ 1 file changed, 4 insertions(+)fs/btrfs/super.c +4
-0@@ -1427,6 +1427,10 @@ static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)seq_puts (seq,",discard=async" );if (!(info->sb->s_flags &SB_POSIXACL ))seq_puts (seq,",noacl" ); +if (info->sb->s_flags &SB_I_VERSION ) +seq_puts (seq,",iversion" ); +else +seq_puts (seq,",noiversion" );if (btrfs_test_opt (info,SPACE_CACHE ))seq_puts (seq,",space_cache" );else if (btrfs_test_opt (info,FREE_SPACE_TREE ))
Quoted 19 lines · Eric Sandeen
Quoted 15 lines · Josef Bacik
Eric reported a problem where if you did
mount -o remount /some/btrfs/fs
you would lose
SB_I_VERSIONon the mountpoint. After a very convoluted search I discovered this is because the remount infrastructure doesn't just say "change these things specifically", but it actually depends on userspace to tell it fucking everything that needs to be set on the mountpoint. This led to the fucking horrifying discovery that util-linux actually has to parse/proc/mountsto figure out what the fuck is set on the mount point in order to preserve any of the options it's not actually fucking with, so in this case iversion. If we don't indicate iversion is set, then we get iversion cleared on the mount, because util-linux doesn't pass inMS_I_VERSIONas it's mount flags.Quoted 2 lines
So work around this fucking insanity by spitting out iversion in
/proc/mountsso we get the correct flags passed to us in remount.
Quoted 7 lines
Hmmm:
# mount -o loop,noiversion btrfsfile mnt # grep btrfs /proc/mounts /dev/loop0 /tmp/mnt btrfs rw,seclabel,relatime,iversion,space_cache,subvolid=5,subvol=/ 0 0 #
Ugh that's because we just set it unconditionally like XFS does, but don't actually pay attention to noiversion otherwise. I'll fix that separately. Thanks,
Josef
Quoted 22 lines · Josef Bacik
Quoted 21 lines · Eric Sandeen
Quoted 15 lines · Josef Bacik
Eric reported a problem where if you did
mount -o remount /some/btrfs/fs
you would lose
SB_I_VERSIONon the mountpoint. After a very convoluted search I discovered this is because the remount infrastructure doesn't just say "change these things specifically", but it actually depends on userspace to tell it fucking everything that needs to be set on the mountpoint. This led to the fucking horrifying discovery that util-linux actually has to parse/proc/mountsto figure out what the fuck is set on the mount point in order to preserve any of the options it's not actually fucking with, so in this case iversion. If we don't indicate iversion is set, then we get iversion cleared on the mount, because util-linux doesn't pass inMS_I_VERSIONas it's mount flags.Quoted 2 lines
So work around this fucking insanity by spitting out iversion in
/proc/mountsso we get the correct flags passed to us in remount.Hmmm:
Quoted 7 lines
Quoted 5 lines
# mount -o loop,noiversion btrfsfile mnt # grep btrfs /proc/mounts /dev/loop0 /tmp/mnt btrfs rw,seclabel,relatime,iversion,space_cache,subvolid=5,subvol=/ 0 0 #Ugh that's because we just set it unconditionally like XFS does, but don't actually pay attention to noiversion otherwise. I'll fix that separately. Thanks,
FWIW,
# mount -o remount,noiversion mnt # grep btrfs /proc/mounts /dev/loop0 /tmp/mnt btrfs rw,seclabel,relatime,noiversion,space_cache,subvolid=5,subvol=/ 0 0
-Eric