[RFC PATCH bpf-next 0/3] Separate error injection framework from kprobes

From: Masami Hiramatsu <mhiramat@kernel.org>
To: Alexei Starovoitov <ast@fb.com>, Josef Bacik <jbacik@fb.com>
Cc: rostedt@goodmis.org, mingo@redhat.com, davem@davemloft.net,
	netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
	ast@kernel.org, kernel-team@fb.com, daniel@iogearbox.net,
	linux-btrfs@vger.kernel.org, darrick.wong@oracle.com,
	mhiramat@kernel.org, Josef Bacik <josef@toxicpanda.com>
Subject: [RFC PATCH bpf-next 0/3] Separate error injection framework from kprobes
Date: Fri, 22 Dec 2017 20:38:13 +0900	[thread overview]
Message-ID: <151394269314.5598.6344820223613246849.stgit@devbox> (raw)

Hi Josef and Alexei,

Here are the patches which describe what I think more "natural"
introduction of error injection APIs. Basically what I did on
this series is to separate error injection from kprobes and put
it on new error-injection small subsystem which is currently
provide whitelists and just-return function stub.

There are 2 main reasons why I separate it from kprobes.

 - kprobes users can modify execution path not only at 
   error-injection whitelist functions but also other
   functions. I don't like to suggest user that such
   limitation is from kprobes itself.

 - This error injection information is also useful for
   ftrace (function-hook) and livepatch. It should not
   be limited by CONFIG_KPROBES.

So I introduced CONFIG_FUNCTION_ERROR_INJECTION for this feature.

This series also have some improvement suggestions.

 - [1/3] "kprobe override function" feature is not limited by
   ftrace-based kprobe, but also you can use it on sw-breakpoint
   based kprobe too. Also, you must check the kprobe is on the
   entry of function right before setting up the stackframe.

 - [2/3] If we store original instruction pointer and compare
   it with regs->ip, we don't need per-cpu bpf_kprobe_override.
   Also, reset_current_kprobe() and preempt_enable_no_resched()
   are no need to separate.

Any thoughts?

If it is good, I also add MAINTAINERS entry for this feature
and add some testcases using kprobes and ftrace to inject
error. (And maybe we also need a document how to use)

BTW, it seems there are many error injection frameworks in
lib/. We may also consider these distinctions.

Thank you,

---

Masami Hiramatsu (3):
      tracing/kprobe: bpf: Check error injectable event is on function entry
      tracing/kprobe: bpf: Compare instruction pointer with original one
      error-injection: Separate error-injection from kprobe

 arch/Kconfig                           |    2 
 arch/x86/Kconfig                       |    2 
 arch/x86/include/asm/error-injection.h |   12 ++
 arch/x86/kernel/kprobes/ftrace.c       |   14 --
 arch/x86/lib/Makefile                  |    2 
 arch/x86/lib/error-inject.c            |   19 +++
 fs/btrfs/disk-io.c                     |    2 
 fs/btrfs/free-space-cache.c            |    2 
 include/asm-generic/error-injection.h  |   20 +++
 include/asm-generic/vmlinux.lds.h      |   14 +-
 include/linux/bpf.h                    |   12 --
 include/linux/error-injection.h        |   21 +++
 include/linux/kprobes.h                |    1 
 include/linux/module.h                 |    6 -
 kernel/kprobes.c                       |  163 --------------------------
 kernel/module.c                        |    8 +
 kernel/trace/Kconfig                   |    4 -
 kernel/trace/bpf_trace.c               |    9 +
 kernel/trace/trace_kprobe.c            |   32 ++---
 kernel/trace/trace_probe.h             |   12 +-
 lib/Kconfig.debug                      |    4 +
 lib/Makefile                           |    1 
 lib/error-inject.c                     |  200 ++++++++++++++++++++++++++++++++
 23 files changed, 323 insertions(+), 239 deletions(-)
 create mode 100644 arch/x86/include/asm/error-injection.h
 create mode 100644 arch/x86/lib/error-inject.c
 create mode 100644 include/asm-generic/error-injection.h
 create mode 100644 include/linux/error-injection.h
 create mode 100644 lib/error-inject.c

--
Signature