On Fri, Jul 17, 2020 at 12:04:18PM -0700, Scott Branden wrote:

On 2020-07-17 10:43 a.m., Kees Cook wrote:

quoted

In preparation for refactoring kernel_read_file*(), remove the redundant
"size" argument which is not needed: it can be included in the return

I don't think the size argument is redundant though.
The existing kernel_read_file functions always read the whole file.
Now, what happens if the file is bigger than the buffer.
How does kernel_read_file know it read the whole file by looking at the
return value?

Yes; an entirely reasonable concern. This is why I add the file_size
output argument later in the series.

quoted

code, with callers adjusted. (VFS reads already cannot be larger than
INT_MAX.)
[...]
-	if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
+	if (i_size > INT_MAX || (max_size > 0 && i_size > max_size)) {

Should this be SSIZE_MAX?

No, for two reasons: then we need to change the return value and likely
the callers need more careful checks, and more importantly, because the
VFS already limits single read actions to INT_MAX, so limits above this
make no sense. Win win! :)

-- 
Kees Cook