Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Florent Revest <revest@chromium.org>
To:     bpf@vger.kernel.org
Cc:     ast@kernel.org, daniel@iogearbox.net, andrii@kernel.org,
        kpsingh@kernel.org, jackmanb@google.com, linux@rasmusvillemoes.dk,
        rostedt@goodmis.org, linux-kernel@vger.kernel.org,
        Florent Revest <revest@chromium.org>
Subject: [PATCH bpf-next v2 0/2] Implement formatted output helpers with bstr_printf
Date:   Tue, 27 Apr 2021 19:43:11 +0200
Message-Id: <20210427174313.860948-1-revest@chromium.org>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

BPF's formatted output helpers are currently implemented with
snprintf-like functions which use variadic arguments. The types of all
arguments need to be known at compilation time. BPF_CAST_FMT_ARG casts
all arguments to the size they should be (known at runtime), but the C
type promotion rules cast them back to u64s. On 32 bit architectures,
this can cause misaligned va_lists and generate mangled output.

This series refactors these helpers to avoid variadic arguments. It uses
a "binary printf" instead, where arguments are passed in a buffer
constructed at runtime.

---
Changes in v2:
- Reworded the second patch's description to better describe how
  arguments get mangled on 32 bit architectures

Florent Revest (2):
  seq_file: Add a seq_bprintf function
  bpf: Implement formatted output helpers with bstr_printf

 fs/seq_file.c            |  18 ++++
 include/linux/bpf.h      |  22 +----
 include/linux/seq_file.h |   4 +
 init/Kconfig             |   1 +
 kernel/bpf/helpers.c     | 188 +++++++++++++++++++++------------------
 kernel/bpf/verifier.c    |   2 +-
 kernel/trace/bpf_trace.c |  34 +++----
 7 files changed, 137 insertions(+), 132 deletions(-)

-- 
2.31.1.498.g6c1eba8ee3d-goog