Re: [PATCH v2 1/5] fsnotify: pass dentry instead of inode when available

[Amir Goldstein <amir73il@gmail.com>]

>
> > 2. exportfs_encode_fh(dentry,... in fanotify_alloc_fid()
> >     Can be converted to exportfs_encode_inode_fh(inode,...
>
> If you have the parent inode then yes. In lot of cases we do have it, not
> sure if in all of them (but likely yes so that we can do proper
> FS_EVENT_ON_CHILD) handling.
>

We do not need the parent inode, because we are not encoding a
"connectable" file handle. we need:
exportfs_encode_inode_fh(inode, (struct fid *)fid->f_handle,
                                  &dwords,  NULL);


> > 3. statfs_by_dentry(dentry,... in fanotify_alloc_fid()
> >     Can be converted to statfs_by_dentry(inode->i_sb->sb_root,...
>
> This might be problematic e.g. with btrfs subvolumes which have each
> different fsid so the fsid-handle pair might be actually invalid or even
> point to a file with different contents. Maybe we could just store the
> fsid in the fsnotify_mark when it is created and use it when generating
> events? That would also get rid of the overhead of calling statfs for each
> generated event which I don't like...
>

OK, I was not being accurate when I wrote that these are the only 3 places
we use dentry. Those are the only 3 places in fanotify, but we also use
dentry->d_sb in fsnotify() to get to the sb mark of course, so we will be using
inode->i_sb instead.

w.r.t btrfs - btrfs has a single sb for multiple subvolumes so by definition the
FAN_MARK_FILESYSTEM feature is only capable of watching ALL subvolumes.

If we wanted to implement subvolume watch for btrfs, we would need
to support attaching a mark to a btrfs subvolume struct (or fs_view [1]).

Basically, the purpose, of fid->fsid is:
1. A key for finding a mount point to use as mount_fd argument to
open_by_handle_at()
2. Make fid unique across the system

Since btrfs file handles encode the subvolume root object id in the file handle,
fid->fsid is only needed to find a btrfs mount of the same sb (blockdev).

Regardless, IIUC, btrfs_statfs() returns an fsid which is associated with the
single super block struct, so all dentries in all subvolumes will
return the same
fsid: btrfs_sb(dentry->d_sb)->fsid.

CC some btrfs folks to correct me if I am wrong.

Thanks,
Amir.

[1] https://lwn.net/Articles/753917/