Re: [PATCH bpf-next v2 1/6] bpf: Introduce pseudo_btf_id

[Andrii Nakryiko]

On Thu, Sep 3, 2020 at 3:34 PM Hao Luo <haoluo@xxxxxxxxxx> wrote:
>
> Pseudo_btf_id is a type of ld_imm insn that associates a btf_id to a
> ksym so that further dereferences on the ksym can use the BTF info
> to validate accesses. Internally, when seeing a pseudo_btf_id ld insn,
> the verifier reads the btf_id stored in the insn[0]'s imm field and
> marks the dst_reg as PTR_TO_BTF_ID. The btf_id points to a VAR_KIND,
> which is encoded in btf_vminux by pahole. If the VAR is not of a struct
> type, the dst reg will be marked as PTR_TO_MEM instead of PTR_TO_BTF_ID
> and the mem_size is resolved to the size of the VAR's type.
>
> From the VAR btf_id, the verifier can also read the address of the
> ksym's corresponding kernel var from kallsyms and use that to fill
> dst_reg.
>
> Therefore, the proper functionality of pseudo_btf_id depends on (1)
> kallsyms and (2) the encoding of kernel global VARs in pahole, which
> should be available since pahole v1.18.
>
> Signed-off-by: Hao Luo <haoluo@xxxxxxxxxx>
> ---

Logic looks correct, but I have a few suggestions for naming things
and verifier logs. Please see below.

>  include/linux/bpf_verifier.h   |   4 ++
>  include/linux/btf.h            |  15 +++++
>  include/uapi/linux/bpf.h       |  38 ++++++++---
>  kernel/bpf/btf.c               |  15 -----
>  kernel/bpf/verifier.c          | 112 ++++++++++++++++++++++++++++++---
>  tools/include/uapi/linux/bpf.h |  38 ++++++++---
>  6 files changed, 182 insertions(+), 40 deletions(-)
>
> diff --git a/include/linux/bpf_verifier.h b/include/linux/bpf_verifier.h
> index 53c7bd568c5d..a14063f64d96 100644
> --- a/include/linux/bpf_verifier.h
> +++ b/include/linux/bpf_verifier.h
> @@ -308,6 +308,10 @@ struct bpf_insn_aux_data {
>                         u32 map_index;          /* index into used_maps[] */
>                         u32 map_off;            /* offset from value base address */
>                 };
> +               struct {
> +                       u32 pseudo_btf_id_type; /* type of pseudo_btf_id */
> +                       u32 pseudo_btf_id_meta; /* memsize or btf_id */

a bit misleading names, not clear at all in the code what's in there.
This section is for ldimm64 insns that are loading BTF variables,
right? so how about this:

struct {
    u32 reg_type;
    union {
        u32 btf_id;
        u32 memsize;
    };
} btf_var;

In case someone hates non-anonymous structs, I'd still go with
btf_var_reg_type, btf_var_btf_id and btf_var_memsize.

> +               };
>         };
>         u64 map_key_state; /* constant (32 bit) key tracking for maps */
>         int ctx_field_size; /* the ctx field size for load insn, maybe 0 */
> diff --git a/include/linux/btf.h b/include/linux/btf.h
> index a9af5e7a7ece..592373d359b9 100644
> --- a/include/linux/btf.h
> +++ b/include/linux/btf.h
> @@ -106,6 +106,21 @@ static inline bool btf_type_is_func_proto(const struct btf_type *t)
>         return BTF_INFO_KIND(t->info) == BTF_KIND_FUNC_PROTO;
>  }
>

[...]

> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index b4e9c56b8b32..3b382c080cfd 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -7398,6 +7398,24 @@ static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn)
>                 return 0;
>         }
>
> +       if (insn->src_reg == BPF_PSEUDO_BTF_ID) {
> +               u32 type = aux->pseudo_btf_id_type;
> +               u32 meta = aux->pseudo_btf_id_meta;
> +
> +               mark_reg_known_zero(env, regs, insn->dst_reg);
> +
> +               regs[insn->dst_reg].type = type;
> +               if (type == PTR_TO_MEM) {
> +                       regs[insn->dst_reg].mem_size = meta;
> +               } else if (type == PTR_TO_BTF_ID) {
> +                       regs[insn->dst_reg].btf_id = meta;

nit: probably worthwhile to introduce a local variable (dst_reg) to
capture pointer to regs[insn->dst_reg] in this entire function. Then
no reall need for type and meta local vars above, everything is going
to be short and sweet.

> +               } else {
> +                       verbose(env, "bpf verifier is misconfigured\n");
> +                       return -EFAULT;
> +               }
> +               return 0;
> +       }
> +
>         map = env->used_maps[aux->map_index];
>         mark_reg_known_zero(env, regs, insn->dst_reg);
>         regs[insn->dst_reg].map_ptr = map;
> @@ -9284,6 +9302,74 @@ static int do_check(struct bpf_verifier_env *env)
>         return 0;
>  }
>
> +/* replace pseudo btf_id with kernel symbol address */
> +static int check_pseudo_btf_id(struct bpf_verifier_env *env,
> +                              struct bpf_insn *insn,
> +                              struct bpf_insn_aux_data *aux)
> +{
> +       u32 type, id = insn->imm;
> +       const struct btf_type *t;
> +       const char *sym_name;
> +       u64 addr;
> +
> +       if (!btf_vmlinux) {
> +               verbose(env, "%s: btf not available. verifier misconfigured.\n",
> +                       __func__);

verifier logs so far hasn't used __func__ and it's not all that
meaningful to users and might change due to later refactorings.

Also, in this particular case, it's not a verifier misconfiguration,
but rather that the kernel doesn't have a built-in BTF, so suggest
enabling CONFIG_DEBUG_INFO_BTF=y.

"kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified
in Kconfig"

> +               return -EINVAL;
> +       }
> +
> +       t = btf_type_by_id(btf_vmlinux, id);
> +       if (!t) {
> +               verbose(env, "%s: invalid btf_id %d\n", __func__, id);

"ldimm64 insn specifies invalid btf_id %d"? add similar context below

> +               return -ENOENT;
> +       }
> +
> +       if (insn[1].imm != 0) {
> +               verbose(env, "%s: BPF_PSEUDO_BTF_ID uses reserved fields\n",
> +                       __func__);
> +               return -EINVAL;
> +       }
> +
> +       if (!btf_type_is_var(t)) {
> +               verbose(env, "%s: btf_id %d isn't KIND_VAR\n", __func__, id);
> +               return -EINVAL;
> +       }
> +
> +       sym_name = btf_name_by_offset(btf_vmlinux, t->name_off);
> +       addr = kallsyms_lookup_name(sym_name);
> +       if (!addr) {
> +               verbose(env, "%s: failed to find the address of symbol '%s'.\n",
> +                       __func__, sym_name);
> +               return -ENOENT;
> +       }
> +

[...]

> @@ -9394,10 +9480,14 @@ static bool bpf_map_is_cgroup_storage(struct bpf_map *map)
>                 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE);
>  }
>
> -/* look for pseudo eBPF instructions that access map FDs and
> - * replace them with actual map pointers
> +/* find and rewrite pseudo imm in ld_imm64 instructions:
> + *
> + * 1. if it accesses map FD, replace it with actual map pointer.
> + * 2. if it accesses btf_id of a VAR, replace it with pointer to the var.
> + *
> + * NOTE: btf_vmlinux is required for converting pseudo btf_id.
>   */
> -static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
> +static int replace_pseudo_imm_with_ptr(struct bpf_verifier_env *env)

nit: I'd call this something like "resolve_pseudo_ldimm64" instead.
ptr here is an implementation detail of map pointers ldimm64 and
doesn't even match what we are doing for pseudo_btf_id

>  {
>         struct bpf_insn *insn = env->prog->insnsi;
>         int insn_cnt = env->prog->len;
> @@ -9438,6 +9528,14 @@ static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env)
>                                 /* valid generic load 64-bit imm */
>                                 goto next_insn;
>
> +                       if (insn[0].src_reg == BPF_PSEUDO_BTF_ID) {
> +                               aux = &env->insn_aux_data[i];
> +                               err = check_pseudo_btf_id(env, insn, aux);
> +                               if (err)
> +                                       return err;
> +                               goto next_insn;
> +                       }
> +
>                         /* In final convert_pseudo_ld_imm64() step, this is
>                          * converted into regular 64-bit imm load insn.
>                          */

[...]