The previous code would actually ignore IO errors during scanning,
becuase s.Scan() returns false if there was an error (leaving the loop
and never being detected). Indeed the correct way[1] of using the
builtin bufio.Scanner is actually more like:

  s := bufio.NewScanner(...)
  for s.Scan() {
      // use s.Text()
  }
  if err := s.Err(); err != nil {
      return err
  }

[1]: https://blog.golang.org/errors-are-values

Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>

Previously, the filter would be run before all of the fields were parsed
(this behaviour also was not documented) -- this resulted in users of
the function accidentally assuming that fields like fsinfo.Root would
actually be filled correctly. It seems that the performance overhead of
parsing a few extra fields is not exorbitant, and optimising this just
leads to incorrect user code.

For a concrete example, this optimisation actually made this runc
change[1] regress a security hardening feature because it relied on
fsinfo.Root being filled correctly.

[1]: opencontainers/runc#2647

Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>

Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>