Jan points out that I forgot to make the needed fixes to the
lz4_uncompress_unknownoutputsize() function to mirror the changes done
in lz4_decompress() with regards to potential pointer overflows.
The only in-kernel user of this function is the zram code, which only
takes data from a valid compressed buffer that it made itself, so it's
not a big issue.  But due to external kernel modules using this
function, it's better to be safe here.
Reported-by: Jan Beulich <JBeulich@suse.com>
Cc: "Don A. Bailey" <donb@securitymouse.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
                        int s = 255;
                        while ((ip < iend) && (s == 255)) {
                                s = *ip++;
+                               if (unlikely(length > (size_t)(length + s)))
+                                       goto _output_error;
                                length += s;
                        }
                }
                if (length == ML_MASK) {
                        while (ip < iend) {
                                int s = *ip++;
+                               if (unlikely(length > (size_t)(length + s)))
+                                       goto _output_error;
                                length += s;
                                if (s == 255)
                                        continue;
 
        /* write overflow error detected */
 _output_error:
-       return (int) (-(((char *) ip) - source));
+       return -1;
 }
 
 int lz4_decompress(const unsigned char *src, size_t *src_len,