Re: [PATCH bpf-next v1 7/9] bpf: Make per_cpu_ptr return rdonly PTR_TO_MEM.

From: Hao Luo <haoluo@google.com>
To: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Cc: Alexei Starovoitov <ast@kernel.org>,
	Andrii Nakryiko <andrii@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Martin KaFai Lau <kafai@fb.com>, Song Liu <songliubraving@fb.com>,
	Yonghong Song <yhs@fb.com>, KP Singh <kpsingh@kernel.org>,
	bpf <bpf@vger.kernel.org>
Subject: Re: [PATCH bpf-next v1 7/9] bpf: Make per_cpu_ptr return rdonly PTR_TO_MEM.
Date: Tue, 7 Dec 2021 19:54:00 -0800	[thread overview]
Message-ID: <CA+khW7jtOweO4it68=ggqbe7QbdhPukE+FkgmAiTs-PeR28AiQ@mail.gmail.com> (raw)
In-Reply-To: <CAEf4BzbSA1+vE4vA6FSbJfUZDyYvyHJbiK1j5yD=vGbGA5EEhg@mail.gmail.com>

On Mon, Dec 6, 2021 at 10:18 PM Andrii Nakryiko
<andrii.nakryiko@gmail.com> wrote:
>
> On Mon, Dec 6, 2021 at 3:22 PM Hao Luo <haoluo@google.com> wrote:
> >
> > Tag the return type of {per, this}_cpu_ptr with RDONLY_MEM. The
> > returned value of this pair of helpers is kernel object, which
> > can not be updated by bpf programs. Previously these two helpers
> > return PTR_OT_MEM for kernel objects of scalar type, which allows
> > one to directly modify the memory. Now with RDONLY_MEM tagging,
> > the verifier will reject programs that writes into RDONLY_MEM.
> >
> > Fixes: 63d9b80dcf2c ("bpf: Introduce bpf_this_cpu_ptr()")
> > Fixes: eaa6bcb71ef6 ("bpf: Introduce bpf_per_cpu_ptr()")
> > Fixes: 4976b718c355 ("bpf: Introduce pseudo_btf_id")
> > Signed-off-by: Hao Luo <haoluo@google.com>
> > ---
> >  kernel/bpf/helpers.c  |  4 ++--
> >  kernel/bpf/verifier.c | 33 ++++++++++++++++++++++++++++-----
> >  2 files changed, 30 insertions(+), 7 deletions(-)
> >
> > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> > index 293d9314ec7f..a5e349c9d3e3 100644
> > --- a/kernel/bpf/helpers.c
> > +++ b/kernel/bpf/helpers.c
> > @@ -667,7 +667,7 @@ BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
> >  const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
> >         .func           = bpf_per_cpu_ptr,
> >         .gpl_only       = false,
> > -       .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL,
> > +       .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY,
> >         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
> >         .arg2_type      = ARG_ANYTHING,
> >  };
> > @@ -680,7 +680,7 @@ BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
> >  const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
> >         .func           = bpf_this_cpu_ptr,
> >         .gpl_only       = false,
> > -       .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID,
> > +       .ret_type       = RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY,
> >         .arg1_type      = ARG_PTR_TO_PERCPU_BTF_ID,
> >  };
> >
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index f8b804918c35..44af65f07a82 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -4296,16 +4296,32 @@ static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regn
> >                                 mark_reg_unknown(env, regs, value_regno);
> >                         }
> >                 }
> > -       } else if (reg->type == PTR_TO_MEM) {
> > +       } else if (base_type(reg->type) == PTR_TO_MEM) {
> > +               bool rdonly_mem = type_is_rdonly_mem(reg->type);
> > +
> > +               if (type_may_be_null(reg->type)) {
> > +                       verbose(env, "R%d invalid mem access '%s'\n", regno,
> > +                               reg_type_str(reg->type));
>
> see, here you'll get "invalid mem access 'ptr_to_mem'" while it's
> actually ptr_to_mem_or_null. Like verifier logs are not hard enough to
> follow, now they will be also misleading.
>

I think formatting string inside reg_type_str() can have this problem
solved, preserving the previous behavior. I'll try that in v2.

> > +                       return -EACCES;
> > +               }
> > +
> > +               if (t == BPF_WRITE && rdonly_mem) {
> > +                       verbose(env, "R%d cannot write into rdonly %s\n",
> > +                               regno, reg_type_str(reg->type));
> > +                       return -EACCES;
> > +               }
> > +
> >                 if (t == BPF_WRITE && value_regno >= 0 &&
> >                     is_pointer_value(env, value_regno)) {
> >                         verbose(env, "R%d leaks addr into mem\n", value_regno);
> >                         return -EACCES;
> >                 }
> > +
> >                 err = check_mem_region_access(env, regno, off, size,
> >                                               reg->mem_size, false);
> > -               if (!err && t == BPF_READ && value_regno >= 0)
> > -                       mark_reg_unknown(env, regs, value_regno);
> > +               if (!err && value_regno >= 0)
> > +                       if (t == BPF_READ || rdonly_mem)
>
> why two nested ifs for one condition?
>

No particular reason. I think it helped me understand the logic
better. But I'm fine with combining them into one 'if'.

> > +                               mark_reg_unknown(env, regs, value_regno);
> >         } else if (reg->type == PTR_TO_CTX) {
> >                 enum bpf_reg_type reg_type = SCALAR_VALUE;
> >                 struct btf *btf = NULL;
>
> [...]