Re: [PATCH] libbpf: Check full backing type size for bitfield dump

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:
- [High] The stricter bounds check using `t->size` incorrectly rejects valid narrow bitfields in packed structs, causing a false positive `-E2BIG` failure.
- [Medium] The bounds check for bitfields was updated in `btf_dump_get_bitfield_value` to use `t->size`, but the early bounds check in `btf_dump_type_data_check_overflow` was left unpatched, causing inconsistent behavior.
- [Low] Unused local variables `start_bit` and `nr_bytes` are left behind after the patch.
--

commit dcf1121c5988139c35dcd52cf31ce52003f38d43
Author: Guangshuo Li <lgs201920130244@gmail.com>

libbpf: Check full backing type size for bitfield dump

This patch updates the bounds check in btf_dump_get_bitfield_value() to use
the full backing type size instead of just the bitfield span. This prevents
out-of-bounds reads when constructing bitfield values from narrow data
buffers.

> diff --git a/tools/lib/bpf/btf_dump.c b/tools/lib/bpf/btf_dump.c
> index cc1ba65bb6c5a..a49790e356a83 100644
> --- a/tools/lib/bpf/btf_dump.c
> +++ b/tools/lib/bpf/btf_dump.c
> @@ -1771,7 +1771,7 @@ static int btf_dump_get_bitfield_value(struct btf_dump *d,
>  	nr_bytes = (start_bit + bit_sz + 7) / 8;

[Severity: Low]
Are start_bit and nr_bytes now unused variables? Since the bounds check
below was updated to use t->size instead of nr_bytes, it looks like these
calculations are dead code and might trigger compiler warnings.

>  
>  	/* Bound check */
> -	if (data + nr_bytes > d->typed_dump->data_end)
> +	if (data + t->size > d->typed_dump->data_end)

[Severity: High]
Will this stricter bounds check reject valid narrow bitfields in packed
structs? If a user dumps a 1-byte packed struct containing an int
bitfield (e.g., int a:8), the backing type size (t->size) will be 4, but
the valid data size is only 1 byte. Could this cause the libbpf BTF dump
API to fail with -E2BIG for valid structs?

[Severity: Medium]
Does this change create an inconsistency with the early bounds check in
btf_dump_type_data_check_overflow()?

Looking at btf_dump_type_data_check_overflow() in the same file:

	if (btf_is_bitfield(t)) {
		...
		return data + nr_bytes > d->typed_dump->data_end ? -E2BIG : nr_bytes;
	}

Since that function still uses nr_bytes, could it succeed while this
function fails, resulting in partial and malformed output where the dump
prints the struct member prefix but fails during value retrieval?

>  		return -E2BIG;
>  
>  	/* Maximum supported bitfield size is 64 bits */

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260705032603.275766-1-lgs201920130244@gmail.com?part=1