On Thu, May 19, 2022 at 7:37 AM Eugene Syromiatnikov <esyr@xxxxxxxxxx> wrote:
>
> On Wed, May 18, 2022 at 04:30:14PM -0700, Andrii Nakryiko wrote:
> > On Tue, May 17, 2022 at 12:36 AM Eugene Syromiatnikov <esyr@xxxxxxxxxx> wrote:
> > >
> > > Check that size would not overflow before calculation (and return
> > > -EOVERFLOW if it will), to prevent potential out-of-bounds write
> > > with the following copy_from_user. Use kvmalloc_array
> > > in copy_user_syms to prevent out-of-bounds write into syms
> > > (and especially buf) as well.
> > >
> > > Fixes: 0dcac272540613d4 ("bpf: Add multi kprobe link")
> > > Cc: <stable@xxxxxxxxxxxxxxx> # 5.18
> > > Signed-off-by: Eugene Syromiatnikov <esyr@xxxxxxxxxx>
> > > ---
> > > kernel/trace/bpf_trace.c | 7 ++++---
> > > 1 file changed, 4 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> > > index 7141ca8..9c041be 100644
> > > --- a/kernel/trace/bpf_trace.c
> > > +++ b/kernel/trace/bpf_trace.c
> > > @@ -2261,11 +2261,11 @@ static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32
> > > int err = -ENOMEM;
> > > unsigned int i;
> > >
> > > - syms = kvmalloc(cnt * sizeof(*syms), GFP_KERNEL);
> > > + syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
> > > if (!syms)
> > > goto error;
> > >
> > > - buf = kvmalloc(cnt * KSYM_NAME_LEN, GFP_KERNEL);
> > > + buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
> > > if (!buf)
> > > goto error;
> > >
> > > @@ -2461,7 +2461,8 @@ int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr
> > > if (!cnt)
> > > return -EINVAL;
> > >
> > > - size = cnt * sizeof(*addrs);
> > > + if (check_mul_overflow(cnt, (u32)sizeof(*addrs), &size))
> > > + return -EOVERFLOW;
> > > addrs = kvmalloc(size, GFP_KERNEL);
> >
> > any good reason not to use kvmalloc_array() here as well and delegate
> > overflow to it. And then use long size (as expected by copy_from_user
> > anyway) everywhere?
>
> Just to avoid double calculation of size, otherwise I don't have
> any significant prefernce, other than -EOVERFLOW would not be reported
> separately (not sure if this a good or a bad thing), and that
> it would be a bit more cumbersome to incorporate the Yonghong's
> suggestion[1] about the INT_MAX check.
>
I think it's totally fine to return ENOMEM if someone requested some
unreasonable amount of symbols. And INT_MAX won't be necessary if we
delegate all the overflow checking to kvmalloc_array()
> [1] https://lore.kernel.org/lkml/412bf136-6a5b-f442-1e84-778697e2b694@xxxxxx/
>
> > > if (!addrs)
> > > return -ENOMEM;
> > > --
> > > 2.1.4
> > >
> >
>
