Re: [Libmtp-discuss] Strange failure with GetPartialObject64 with seek past EOF

[Philip L. <ph...@ov...>]

On Thu, 16 Nov 2017 14:03:26 -0800
Jerry Zhang <zha...@go...> wrote:

> Hi all,
> 
> It looks like the behavior would be caused by this line in the kernel
> mtp driver. The offset itself is not sanitized before being passed to
> 
> ret = vfs_read(filp, req->buf + hdr_size, xfer - hdr_size, &offset);
> 
> vfs_read does not perform the same checks as a read() syscall, which
> would explain the weird behavior.
> 
> However, we're switching over to using a functionfs interface instead.
> Currently, only Pixel and Pixel 2 phones have made the switch since it
> is up to oems to do it themselves. So for these phones, we have
> 
>  if (TEMP_FAILURE_RETRY(pread(mfr.fd, mIobuf[0].bufs.data() +
>                  sizeof(mtp_data_header), init_read_len, offset))
>          != init_read_len) return -1;
> 
> Since this uses pread, which will have posix semantics, it makes sense
> that pixel and pixel 2 will not have this error. Any other phone, no
> matter what OS, will probably run into this error so its a good idea
> to have workaround for backwards compatibility.
> 
> Thanks,
> Jerry

Hi Jerry,

Thanks for the explanation - the only pixel device in our tested set
was mine, so it all lines up.

We'll go ahead and apply the workaround.

--phil