Re: [PATCH bpf-next v9 2/2] selftests/bpf: Check bpftool batch map dumps

[Quentin Monnet <qmo@kernel.org>]

On 04/11/2026 19:54, Tianyi Chen wrote:
> Batched dumps must print the same output as individual lookups. Use an
> fmod_ret program on security_bpf() to count the map lookups made by
> bpftool children of the test, and to fail batch lookups with -EBUSY.
> 
> For each map type with batched lookups, compare plain and JSON dumps
> with a dump whose batch lookups all fail, and check that no individual
> lookups were needed. Cover BTF and per-CPU output, an empty map, 1-byte
> keys and unaligned values. With 4 KiB values, check dumps that need
> several batches, and that failing the first or second batch lookup
> continues with individual lookups without restarting the dump.
> 
> Pre-size RHASH maps so asynchronous growth does not race with the
> order-sensitive dump comparisons.
> 
> Assisted-by: LLM
> Signed-off-by: Tianyi Chen <hi@tychen.cc>


Acked-by: Quentin Monnet <qmo@kernel.org>

... with a few, non-blocking suggestions, inline below.


> ---
>  .../bpf/prog_tests/bpftool_map_batch.c        | 231 ++++++++++++++++++
>  .../selftests/bpf/progs/bpftool_map_batch.c   |  42 ++++
>  2 files changed, 273 insertions(+)
>  create mode 100644 tools/testing/selftests/bpf/prog_tests/bpftool_map_batch.c
>  create mode 100644 tools/testing/selftests/bpf/progs/bpftool_map_batch.c
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/bpftool_map_batch.c b/tools/testing/selftests/bpf/prog_tests/bpftool_map_batch.c
> new file mode 100644
> index 0000000000000..4cca77f410a06
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/prog_tests/bpftool_map_batch.c


Maybe bpftool_map_dump_batch.c, so it's easier to visually group with
the existing bpftool_map_dump.c?


> @@ -0,0 +1,231 @@

[...]

> +void test_bpftool_map_batch(void)
> +{
> +	static const struct dump_case cases[] = {
> +		{ "hash", BPF_MAP_TYPE_HASH, 4, 4, 100 },
> +		{ "empty", BPF_MAP_TYPE_HASH, 4, 4, 0 },
> +		/* The hash cursor is wider than the key; values are unaligned. */
> +		{ "odd_sizes", BPF_MAP_TYPE_HASH, 1, 3, 100 },
> +		{ "btf", BPF_MAP_TYPE_HASH, 4, 4, 100, true },
> +		{ "lru_hash", BPF_MAP_TYPE_LRU_HASH, 4, 4, 100 },
> +		{ "percpu_hash", BPF_MAP_TYPE_PERCPU_HASH, 4, 3, 100 },
> +		{ "lru_percpu_hash", BPF_MAP_TYPE_LRU_PERCPU_HASH, 4, 4, 100 },


If I remember correctly, lru_hash/lru_percpu_hash are mostly the same
code as hash/percpu_hash, so I'm not sure they're worth testing, unless
you want completeness.

You could also avoid checking both plain and JSON for all types (pick
one, and just do a few check for the other?), if you want to save some
bpftool invocations; I don't expect differences in the code related to
batch lookups between the plain and JSON output printing code.


> +		{ "btf_percpu_hash", BPF_MAP_TYPE_PERCPU_HASH, 4, 4, 100, true },
> +		{ "rhash", BPF_MAP_TYPE_RHASH, 4, 4, 100 },
> +		{ "array", BPF_MAP_TYPE_ARRAY, 4, 4, 100 },
> +		{ "percpu_array", BPF_MAP_TYPE_PERCPU_ARRAY, 4, 3, 100 },
> +		{ "lpm_trie", BPF_MAP_TYPE_LPM_TRIE, 8, 4, 100 },
> +		{ "array_of_maps", BPF_MAP_TYPE_ARRAY_OF_MAPS, 4, 4, 100 },
> +		{ "hash_of_maps", BPF_MAP_TYPE_HASH_OF_MAPS, 4, 4, 100 },
> +		/* Bucket and key cursors, up to 255 entries per batch. */
> +		{ "multi_batch_hash", BPF_MAP_TYPE_HASH, 4, 4096, 256, false, true },
> +		{ "multi_batch_array", BPF_MAP_TYPE_ARRAY, 4, 4096, 256, false, true },
> +	};
> +	struct bpftool_map_batch *skel;
> +	char *expected = NULL, *output = NULL;
> +	unsigned int i;
> +
> +	skel = bpftool_map_batch__open_and_load();
> +	if (!ASSERT_OK_PTR(skel, "skel_open_and_load"))
> +		return;
> +	skel->bss->test_tgid = getpid();
> +	if (!ASSERT_OK(bpftool_map_batch__attach(skel), "skel_attach"))
> +		goto out;
> +	expected = malloc(OUTPUT_SIZE);
> +	output = malloc(OUTPUT_SIZE);
> +	if (!ASSERT_OK_PTR(expected, "expected") || !ASSERT_OK_PTR(output, "output"))
> +		goto out;
> +
> +	for (i = 0; i < ARRAY_SIZE(cases); i++)
> +		if (test__start_subtest(cases[i].name))
> +			run_case(skel, &cases[i], expected, output);
> +out:
> +	free(expected);
> +	free(output);
> +	bpftool_map_batch__destroy(skel);
> +}
> diff --git a/tools/testing/selftests/bpf/progs/bpftool_map_batch.c b/tools/testing/selftests/bpf/progs/bpftool_map_batch.c
> new file mode 100644
> index 0000000000000..0ca6f7973e880
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/progs/bpftool_map_batch.c

bpftool_map_dump_batch.c?