On Thu, Dec 9, 2021 at 12:04 PM Hao Luo <haoluo@xxxxxxxxxx> wrote:
>
> On Tue, Dec 7, 2021 at 7:49 PM Hao Luo <haoluo@xxxxxxxxxx> wrote:
> >
> > On Mon, Dec 6, 2021 at 10:24 PM Andrii Nakryiko
> > <andrii.nakryiko@xxxxxxxxx> wrote:
> > >
> > > On Mon, Dec 6, 2021 at 3:22 PM Hao Luo <haoluo@xxxxxxxxxx> wrote:
> > > >
> > > > Some helper functions may modify its arguments, for example,
> > > > bpf_d_path, bpf_get_stack etc. Previously, their argument types
> > > > were marked as ARG_PTR_TO_MEM, which is compatible with read-only
> > > > mem types, such as PTR_TO_RDONLY_BUF. Therefore it's legitimate
> > > > to modify a read-only memory by passing it into one of such helper
> > > > functions.
> > > >
> > > > This patch tags the bpf_args compatible with immutable memory with
> > > > MEM_RDONLY flag. The arguments that don't have this flag will be
> > > > only compatible with mutable memory types, preventing the helper
> > > > from modifying a read-only memory. The bpf_args that have
> > > > MEM_RDONLY are compatible with both mutable memory and immutable
> > > > memory.
> > > >
> > > > Signed-off-by: Hao Luo <haoluo@xxxxxxxxxx>
> > > > ---
> > > > include/linux/bpf.h | 4 ++-
> > > > kernel/bpf/btf.c | 2 +-
> > > > kernel/bpf/cgroup.c | 2 +-
> > > > kernel/bpf/helpers.c | 8 ++---
> > > > kernel/bpf/ringbuf.c | 2 +-
> > > > kernel/bpf/syscall.c | 2 +-
> > > > kernel/bpf/verifier.c | 14 +++++++--
> > > > kernel/trace/bpf_trace.c | 26 ++++++++--------
> > > > net/core/filter.c | 64 ++++++++++++++++++++--------------------
> > > > 9 files changed, 67 insertions(+), 57 deletions(-)
> > > >
> [...]
> > > > @@ -5074,6 +5074,7 @@ static int check_reg_type(struct bpf_verifier_env *env, u32 regno,
> > > > struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno];
> > > > enum bpf_reg_type expected, type = reg->type;
> > > > const struct bpf_reg_types *compatible;
> > > > + u32 compatible_flags;
> > > > int i, j;
> > > >
> > > > compatible = compatible_reg_types[base_type(arg_type)];
> > > > @@ -5082,6 +5083,13 @@ static int check_reg_type(struct bpf_verifier_env *env, u32 regno,
> > > > return -EFAULT;
> > > > }
> > > >
> > > > + /* If arg_type is tagged with MEM_RDONLY, it's compatible with both
> > > > + * RDONLY and non-RDONLY reg values. Therefore fold this flag before
> > > > + * comparison. PTR_MAYBE_NULL is similar.
> > > > + */
> > > > + compatible_flags = arg_type & (MEM_RDONLY | PTR_MAYBE_NULL);
> > > > + type &= ~compatible_flags;
> > > > +
> > >
> > > wouldn't
> > >
> > > type &= ~MEM_RDONLY; /* clear read-only flag, if any */
> > > type &= ~PTR_MAYBE_NULL; /* clear nullable flag, if any */
> > >
> > > be cleaner and more straightforward?
> > >
> > >
> >
> > No problem. Sounds good to me.
> >
>
> I just realized the suggested transformation is wrong. Whether to fold
> the flag depends on whether arg_type has the flag. So it should
> instead be
>
> if (arg_type & MEM_RDONLY)
> type &= ~MEM_RDONLY;
>
> or
>
> type &= ~(arg_type & MEM_RDONLY);
You are totally right. I think this deserves a big verbose comment
explaining that:
ARG_PTR_TO_MEM+RDONLY is compatible with PTR_TO_MEM and PTR_TO_MEM+RDONLY, but
ARG_PTR_TO_MEM is compatible only with PTR_TO_MEM and NOT with PTR_TO_MEM+RDONLY
Same for MAYBE_NULL:
ARG_PTR_TO_MEM + MAYBE_NULL is compatible with PTR_TO_MEM and
PTR_TO_MEM+MAYBE_NULL, but
ARG_PTR_TO_MEM is compatible only with PTR_TO_MEM but NOT with
PTR_TO_MEM+MAYBE_NULL
It might not be true for other future modifiers, so I'd do each of
RDONLY and MAYBE_NULL with a separate if and comment.
Good catch, btw! (but hopefully selftests would have caught this? if
not, we need better tests)
>
> > > > for (i = 0; i < ARRAY_SIZE(compatible->types); i++) {
> > > > expected = compatible->types[i];
> > > > if (expected == NOT_INIT)
> > >
> > > [...]
