On Tue, Nov 30, 2021 at 11:19 PM Hou Tao <houtao1@xxxxxxxxxx> wrote:
>
> BPF_LOG_KERNEL is only used internally, so disallow bpf_btf_load()
> to set log level as BPF_LOG_KERNEL. The same checking has already
> been done in bpf_check(), so factor out a helper to check the
> validity of log attributes and use it in both places.
>
> Fixes: 8580ac9404f6 ("bpf: Process in-kernel BTF")
> Signed-off-by: Hou Tao <houtao1@xxxxxxxxxx>
> Acked-by: Yonghong Song <yhs@xxxxxx>
> Acked-by: Martin KaFai Lau <kafai@xxxxxx>
> ---
> include/linux/bpf_verifier.h | 7 +++++++
> kernel/bpf/btf.c | 3 +--
> kernel/bpf/verifier.c | 6 +++---
> 3 files changed, 11 insertions(+), 5 deletions(-)
>
> diff --git a/include/linux/bpf_verifier.h b/include/linux/bpf_verifier.h
> index c8a78e830fca..a3d17601a5a7 100644
> --- a/include/linux/bpf_verifier.h
> +++ b/include/linux/bpf_verifier.h
> @@ -396,6 +396,13 @@ static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log)
> log->level == BPF_LOG_KERNEL);
> }
>
> +static inline bool
> +bpf_verifier_log_attr_valid(const struct bpf_verifier_log *log, u32 max_total)
> +{
> + return log->len_total >= 128 && log->len_total <= max_total &&
> + log->level && log->ubuf && !(log->level & ~BPF_LOG_MASK);
> +}
> +
> #define BPF_MAX_SUBPROGS 256
>
> struct bpf_subprog_info {
> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
> index 6b9d23be1e99..308c345cd811 100644
> --- a/kernel/bpf/btf.c
> +++ b/kernel/bpf/btf.c
> @@ -4472,8 +4472,7 @@ static struct btf *btf_parse(bpfptr_t btf_data, u32 btf_data_size,
> log->len_total = log_size;
>
> /* log attributes have to be sane */
> - if (log->len_total < 128 || log->len_total > UINT_MAX >> 8 ||
> - !log->level || !log->ubuf) {
> + if (!bpf_verifier_log_attr_valid(log, UINT_MAX >> 8)) {
> err = -EINVAL;
> goto errout;
> }
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 722aea00d44e..f128e6799cb5 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -13969,11 +13969,11 @@ int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, bpfptr_t uattr)
> log->ubuf = (char __user *) (unsigned long) attr->log_buf;
> log->len_total = attr->log_size;
>
> - ret = -EINVAL;
> /* log attributes have to be sane */
> - if (log->len_total < 128 || log->len_total > UINT_MAX >> 2 ||
> - !log->level || !log->ubuf || log->level & ~BPF_LOG_MASK)
> + if (!bpf_verifier_log_attr_valid(log, UINT_MAX >> 2)) {
> + ret = -EINVAL;
It's actually quite bad that we have this discrepancy in limits.
I've already sent a patch to make them the same.
It was a pain to debug.
https://lore.kernel.org/bpf/20211124060209.493-7-alexei.starovoitov@xxxxxxxxx/
"
Otherwise tools that progressively increase log size and use the same log
for BTF loading and program loading will be hitting hard to debug EINVAL.
"
