Re: [fuse-devel] copy_file_range return value on FUSE

[Florian Weimer <fweimer@redhat.com>]

* Miklos Szeredi:

> On Mon, 4 Aug 2025 at 11:42, Florian Weimer via fuse-devel
> <fuse-devel@lists.sourceforge.net> wrote:
>>
>> The FUSE protocol uses struct fuse_write_out to convey the return value
>> of copy_file_range, which is restricted to uint32_t.  But the
>> copy_file_range interface supports a 64-bit copy operation.  Given that
>> copy_file_range is expected to clone huge files, large copies are not
>> unexpected, so this appears to be a real limitation.
>
> That's a nasty oversight.  Fixing with a new FUSE_COPY_FILE_RANGE_64
> op, fallback to the legacy FUSE_COPY_FILE_RANGE.

Or adding a capability flag to switch from struct fuse_write_out to
something that uses an uint64_t value.  One complication: The struct
fuse_write_out layout is too close to a potential 64-bit version of it
on little-endian systems, so that proper testing might be difficult with
the obvious approach.

>> There is another wrinkle: we'd need to check if the process runs in
>> 32-bit compat mode, and reject size_t arguments larger than INT_MAX in
>> this case (with EOVERFLOW presumably).  But perhaps this should be
>> handled on the kernel side?  Currently, this doesn't seem to happen, and
>> we can get copy_file_range results in the in-band error range.
>> Applications have no way to disambiguate this.
>
> That's not fuse specific, right?

In-kernel file systems can check if the request originated from a compat
process, using in_compat_syscall.  I don't think that's possible over
FUSE.

Thanks,
Florian