Re: [PATCH bpf-next v1 2/8] bpf: Add bpf_dynptr_trim and bpf_dynptr_advance

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Wed, Sep 7, 2022 at 5:10 PM Joanne Koong <joannelkoong@xxxxxxxxx> wrote:
>
> Add two new helper functions: bpf_dynptr_trim and bpf_dynptr_advance.
>
> bpf_dynptr_trim decreases the size of a dynptr by the specified
> number of bytes (offset remains the same). bpf_dynptr_advance advances
> the offset of the dynptr by the specified number of bytes (size
> decreases correspondingly).
>
> One example where trimming / advancing the dynptr may useful is for
> hashing. If the dynptr points to a larger struct, it is possible to hash
> an individual field within the struct through dynptrs by using
> bpf_dynptr_advance+trim.
>
> Signed-off-by: Joanne Koong <joannelkoong@xxxxxxxxx>
> ---
>  include/uapi/linux/bpf.h       | 16 +++++++++
>  kernel/bpf/helpers.c           | 63 ++++++++++++++++++++++++++++++++++
>  tools/include/uapi/linux/bpf.h | 16 +++++++++
>  3 files changed, 95 insertions(+)
>
> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> index cce3356765fc..3b054553be30 100644
> --- a/include/uapi/linux/bpf.h
> +++ b/include/uapi/linux/bpf.h
> @@ -5453,6 +5453,20 @@ union bpf_attr {
>   *             dynptr is invalid or if the offset and length is out of bounds
>   *             or in a paged buffer for skb-type dynptrs or across fragments
>   *             for xdp-type dynptrs.
> + *
> + * long bpf_dynptr_advance(struct bpf_dynptr *ptr, u32 len)
> + *     Description
> + *             Advance a dynptr by *len*.

This might be a bit to succinct a description. "Advance dynptr's
internal offset by *len* bytes."?

> + *     Return
> + *             0 on success, -EINVAL if the dynptr is invalid, -ERANGE if *len*
> + *             exceeds the bounds of the dynptr.
> + *
> + * long bpf_dynptr_trim(struct bpf_dynptr *ptr, u32 len)
> + *     Description
> + *             Trim a dynptr by *len*.

Similar as above, something like "Reduce the size of memory pointed to
by dynptr by *len* without changing offset"?

> + *     Return
> + *             0 on success, -EINVAL if the dynptr is invalid, -ERANGE if
> + *             trying to trim more bytes than the size of the dynptr.
>   */
>  #define __BPF_FUNC_MAPPER(FN)          \
>         FN(unspec),                     \
> @@ -5667,6 +5681,8 @@ union bpf_attr {
>         FN(dynptr_from_skb),            \
>         FN(dynptr_from_xdp),            \
>         FN(dynptr_data_rdonly),         \
> +       FN(dynptr_advance),             \
> +       FN(dynptr_trim),                \
>         /* */
>
>  /* integer value in 'imm' field of BPF_CALL instruction selects which helper
> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> index 30a59c9e5df3..9f356105ab49 100644
> --- a/kernel/bpf/helpers.c
> +++ b/kernel/bpf/helpers.c
> @@ -1423,6 +1423,13 @@ static u32 bpf_dynptr_get_size(struct bpf_dynptr_kern *ptr)
>         return ptr->size & DYNPTR_SIZE_MASK;
>  }
>
> +static void bpf_dynptr_set_size(struct bpf_dynptr_kern *ptr, u32 new_size)
> +{
> +       u32 metadata = ptr->size & ~DYNPTR_SIZE_MASK;
> +
> +       ptr->size = new_size | metadata;
> +}
> +
>  int bpf_dynptr_check_size(u32 size)
>  {
>         return size > DYNPTR_MAX_SIZE ? -E2BIG : 0;
> @@ -1646,6 +1653,58 @@ static const struct bpf_func_proto bpf_dynptr_data_proto = {
>         .arg3_type      = ARG_CONST_ALLOC_SIZE_OR_ZERO,
>  };
>
> +BPF_CALL_2(bpf_dynptr_advance, struct bpf_dynptr_kern *, ptr, u32, len)
> +{
> +       u32 size;
> +
> +       if (!ptr->data)
> +               return -EINVAL;
> +
> +       size = bpf_dynptr_get_size(ptr);
> +
> +       if (len > size)
> +               return -ERANGE;
> +
> +       ptr->offset += len;
> +
> +       bpf_dynptr_set_size(ptr, size - len);

Kind of weird that offset is adjusted directly through field, but size
is set through helper.

One idea is to have (internal) helper that *increments* offset and
*decrements* size as one step. You can move all the error checking
inside it, instead of duplicating. E.g., something like this signature

static int bpf_dynptr_adjust(struct btf_dynptr_kern *ptr, u32 off_inc,
u32 sz_dec);

This increment/decrement is a bit surprising, but we don't allow size
to grow and offset to decrement, so this makes sense from that point
of view.

With that, we'll just do

return bpf_dynptr_adjust(ptr, len, len);

> +
> +       return 0;
> +}
> +
> +static const struct bpf_func_proto bpf_dynptr_advance_proto = {
> +       .func           = bpf_dynptr_advance,
> +       .gpl_only       = false,
> +       .ret_type       = RET_INTEGER,
> +       .arg1_type      = ARG_PTR_TO_DYNPTR,
> +       .arg2_type      = ARG_ANYTHING,
> +};
> +
> +BPF_CALL_2(bpf_dynptr_trim, struct bpf_dynptr_kern *, ptr, u32, len)
> +{
> +       u32 size;
> +
> +       if (!ptr->data)
> +               return -EINVAL;
> +
> +       size = bpf_dynptr_get_size(ptr);
> +
> +       if (len > size)
> +               return -ERANGE;
> +
> +       bpf_dynptr_set_size(ptr, size - len);

and here we can just do

return bpf_dynptr_adjust(ptr, 0, len);

> +
> +       return 0;
> +}
> +
> +static const struct bpf_func_proto bpf_dynptr_trim_proto = {
> +       .func           = bpf_dynptr_trim,
> +       .gpl_only       = false,
> +       .ret_type       = RET_INTEGER,
> +       .arg1_type      = ARG_PTR_TO_DYNPTR,
> +       .arg2_type      = ARG_ANYTHING,
> +};

[...]



[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux