Re: [BUG] kernel: kcov: a possible sleep-in-atomic-context bug in kcov_ioctl()

[Dmitry Vyukov]

On Tue, Dec 17, 2019 at 2:11 PM Jia-Ju Bai <baijiaju1990@xxxxxxxxx> wrote:
> On 2019/12/17 21:02, Dmitry Vyukov wrote:
> > On Tue, Dec 17, 2019 at 1:56 PM Jia-Ju Bai <baijiaju1990@xxxxxxxxx> wrote:
> >> The kernel may sleep while holding a spinlock.
> >> The function call path (from bottom to top) in Linux 4.19 is:
> >>
> >> kernel/kcov.c, 237:
> >>       vfree in kcov_put
> >> kernel/kcov.c, 413:
> >>       kcov_put in kcov_ioctl_locked
> >> kernel/kcov.c, 427:
> >>       kcov_ioctl_locked in kcov_ioctl
> >> kernel/kcov.c, 426:
> >>       spin_lock in kcov_ioctl
> >>
> >> vfree() can sleep at runtime.
> >>
> >> I am not sure how to properly fix this possible bug, so I only report it.
> >> A possible way is to replace vfree() with kfree(), and replace related
> >> calls to vmalloc() with kmalloc().
> >>
> >> This bug is found by a static analysis tool STCheck written by myself.
> > Hi Jia-Ju,
> >
> > Are you sure kcov_ioctl_locked can really release the descriptor? It
> > happens in the context of ioctl, which means there is an open
> > reference for the file descriptor. So ioctl should not do vfree I
> > would assume.
>
> Thanks for the reply :)
> I am not sure, because I am not familiar with kcov.
> But looking at the code, if the reference count of kcov is 1, vfree()
> could be called.

That kcov_put should never call vfree. We still hold reference
associated with the file, which will be released in kcov_close.