Re: [fuse-devel] fuse_lowlevel_ops::getattr( ..., struct fuse_file_info* fi != NULL)

[Miklos S. <mi...@sz...>]

On Wed, Apr 29, 2015 at 2:48 PM, John Salmon
<Joh...@de...> wrote:
>
> The documentation says that the fi argument to the lowlevel getattr
> callback is "for future use, currently always NULL".
>
> But in practice, I see that it is sometimes non-NULL.  Is there
> documentation describing when it is non-NULL.  In the cases I've seen,
> it is non-NULL when getattr is called to verify EOF on an open file.
> But I'm curious about the details and whether there are other
> situations where it is non-NULL.

It's used from read, write and lseek to verify EOF.

> When I noticed that fi was sometimes non-NULL, I had high hopes that
> it would be non-NULL when a process called fstat on an open file
> descriptor, and that fi->fh would contain the value set by the
> corresponding open.  Unfortunately, this seems not to be the case.
> The behavior seems to depend on whether whether read was called
> between the open and the fstat.  For example, this idiom (which seems
> to be common in dlopen) results in a call to
> lowlevel_ops::getattr(..., fi=NULL).
>
>     fd = open("/file/managed/by/my/fuse_process", O_RDONLY);
>     read(fd, buf, sz);
>     fstat(fd, &sb);
>
> Is there any way I can determine, in the fuse daemon, that the getattr
> resulting from the above fstat is associated with a currently open
> file?

No.  The linux VFS doesn't provide that information to filesystems, so
fuse can't provide it to userspace filesystems.

Thanks,
Miklos