Re: [PATCH] tracing: Correct the length check which causes memory corruption

[Linus Torvalds]

On Wed, Jun 9, 2021 at 1:52 PM Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> >
> > That "sizeof(*entry)" is clearly wrong, because it doesn't take the
> > unsized array into account.
>
> Correct. That's because I forgot that the structure has that empty array :-(

Note that 'sparse' does have the option to warn about odd flexible
array uses. Including 'sizeof()'.

You can do something like

    CF='-Wflexible-array-sizeof' make C=2 kernel/trace/trace.o

and you'll see

  kernel/trace/trace.c:1022:17: warning: using sizeof on a flexible structure
  kernel/trace/trace.c:2645:17: warning: using sizeof on a flexible structure
  kernel/trace/trace.c:2739:41: warning: using sizeof on a flexible structure
  kernel/trace/trace.c:3290:16: warning: using sizeof on a flexible structure
  kernel/trace/trace.c:3350:16: warning: using sizeof on a flexible structure
  kernel/trace/trace.c:6989:16: warning: using sizeof on a flexible structure
  kernel/trace/trace.c:7070:16: warning: using sizeof on a flexible structure

and I suspect every single one of those should be using
'struct_size()' instead for a sizeof() on the base structure plus some
manual arithmetic (or, as in the case of this bug, _without_ the extra
arithmetic).

And yeah, it isn't just the tracing code that does this. We have it
all over, so that sparse check isn't on by default. Sparse is pretty
darn noisy even without it, but it can be worth using that
CF='-Wflexible-array-sizeof' on individual files that you want to
check.

               Linus