Re: kernel BUG at lib/string.c:LINE! (6)

[Dmitry Vyukov]

On Tue, Dec 22, 2020 at 11:07 PM Florian Westphal <fw@xxxxxxxxx> wrote:
>
> Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> > On Tue, Dec 22, 2020 at 6:44 AM syzbot
> > <syzbot+e86f7c428c8c50db65b4@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > The issue was bisected to:
> > >
> > > commit 2f78788b55ba ("ilog2: improve ilog2 for constant arguments")
> >
> > That looks unlikely, although possibly some constant folding
> > improvement might make the fortify code notice something with it.
> >
> > > detected buffer overflow in strlen
> > > ------------[ cut here ]------------
> > > kernel BUG at lib/string.c:1149!
> > > Call Trace:
> > >  strlen include/linux/string.h:325 [inline]
> > >  strlcpy include/linux/string.h:348 [inline]
> > >  xt_rateest_tg_checkentry+0x2a5/0x6b0 net/netfilter/xt_RATEEST.c:143
> >
> > Honestly, this just looks like the traditional bug in "strlcpy()".
>
> Yes, thats exactly what this is, no idea why the bisection points
> at ilog2 changes.

The end result is usually clear from the bisection log:

> bisection log:  https://syzkaller.appspot.com/x/bisect.txt?x=1584f137500000

In this case it looks like the most common cause of diverted bisection
-- interference from other kernel bugs, this __queue_work issue that
happened on ilog2 commit:

[03f4935135b9efeb780b970ba023c201f81cf4e6] checkpatch: fix unescaped left brace
testing commit 03f4935135b9efeb780b970ba023c201f81cf4e6 with gcc (GCC) 8.1.0
all runs: crashed: kernel BUG at lib/string.c:LINE!
# git bisect bad 03f4935135b9efeb780b970ba023c201f81cf4e6

Bisecting: 21 revisions left to test after this (roughly 5 steps)
[2f78788b55baa3410b1ec91a576286abe1ad4d6a] ilog2: improve ilog2 for
constant arguments
testing commit 2f78788b55baa3410b1ec91a576286abe1ad4d6a with gcc (GCC) 8.1.0
run #0: crashed: WARNING in __queue_work
# git bisect bad 2f78788b55baa3410b1ec91a576286abe1ad4d6a




> > That BSD function is complete garbage, exactly because it doesn't
> > limit the source length. People tend to _think_ it does ("what's that
> > size_t argument for?") but strlcpy() only limits the *destination*
> > size, and the source is always read fully.
>
> Right, I'll send a patch shortly.