On Mon, Jun 10, 2019 at 11:39 PM Steve French <smfrench@xxxxxxxxx> wrote:
>
> Looks good in my testing so far - but also want to do a little more
> testing with the copy_file_range xfstest cases because your patches
> fixed one additional test (not cross mount copy) so we can understand
> why it fixed that test case.
I know which of my patches fixed generic/43[01].
It was 96e6e8f4a68d ("vfs: add missing checks to copy_file_range")
More specifically this code:
/* Shorten the copy to EOF */
size_in = i_size_read(inode_in);
if (pos_in >= size_in)
count = 0;
else
count = min(count, size_in - (uint64_t)pos_in);
If CIFS sends an out of range value of copy length to Windows server,
server replies with an error. That is inconsistent with the semantics of
copy_file_range(2) syscall, which expects "short copy", hence need
to shorten length before passing on to server.
I verified with Aurelien that this is the case and I was under the impression
that he was going to create a similar local fix to cifs code for stable.
I thought he told you, so I forgot to report back myself.
Thanks,
Amir.
