Skip to content

bpf: Tracing recursion prevention mechanism improvement

Yafang Shao bpf linux-trace-kernel lkml oe-kbuild-all 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest6 patches 17 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/6bpf: Add __rcu_read_{lock,unlock} into btf id deny listkernel/bpf/verifier.c+4 -00 replies
2/6tracing: Add generic test_recursion_try_acquire()include/linux/trace_recursion.h kernel/trace/ftrace.c+37 -121 reply
3/6tracing: Add the comment for allowing one single recursion in process contextinclude/linux/trace_recursion.h+2 -00 replies
4/6selftests/bpf: Allow one single recursion in fentry recursion testtools/testing/selftests/bpf/prog_tests/recursion.c tools/testing/selftests/bpf/test_progs.h+24 -20 replies
5/6bpf: Improve tracing recursion prevention mechanisminclude/linux/bpf.h kernel/bpf/core.c kernel/bpf/trampoline.c +1 more+41 -2716 replies
6/6bpf: Remove some denied functions from the btf id deny listkernel/bpf/verifier.c+0 -80 replies

Reviews trailers per patch

1 of 6 patches has a Reviewed-by or Acked-by.

PatchYafang ShaoMasami Hiramatsu (Google)
1/6
2/6A-b R-b
3/6
4/6
5/6
6/6

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.

Read the v1 thread

Cover letter [PATCH bpf-next 0/6] bpf: Tracing recursion prevention mechanism improvement

Currently we use prog->active to prevent tracing recursion, but it has some downsides,

- It can't identify different contexts
That said, if a process context is interrupted by a irq context and the irq context runs the same code path, it will be considered as recursion. For example,

    normal:
      this_cpu_inc_return(*(prog->active)) == 1 <- OK

      irq:
        this_cpu_inc_return(*(prog->active)) == 1 <- FAIL!
        [ Considered as recusion ]

- It has to maintain a percpu area
A percpu area will be allocated for each prog when the prog is loaded and be freed when the prog is destroyed.

Let's replace it with the generic tracing recursion prevention mechanism, which can work fine with anything. In the above example, the irq context won't be considered as recursion again,
normal:
test_recursion_try_acquire() <- OK

softirq:
test_recursion_try_acquire() <- OK

irq:
test_recursion_try_acquire() <- OK

Note that, currently one single recursion in process context is allowed due to the TRACE_CTX_TRANSITION workaround, which can be fixed in the future. That said, below behavior is expected currently,
normal:
test_recursion_try_acquire() <- OK
[ recursion happens ] <- one single recursion is allowed
test_recursion_try_acquire() <- OK
[ recursion happens ]
test_recursion_try_acquire() <- RECURSION!

To adapt to this behavior, the bpf recursion selftest is changed.

Steven Rostedt (Google) (1):
tracing: Add generic test_recursion_try_acquire()

Yafang Shao (5):
bpf: Add __rcu_read_{lock,unlock} into btf id deny list
tracing: Add the comment for allowing one single recursion in process
context
selftests/bpf: Allow one single recursion in fentry recursion test
bpf: Improve tracing recursion prevention mechanism
bpf: Remove some denied functions from the btf id deny list

 include/linux/bpf.h                                |  2 +-
 include/linux/trace_recursion.h                    | 49 ++++++++++++++++------
 kernel/bpf/core.c                                  | 10 -----
 kernel/bpf/trampoline.c                            | 44 ++++++++++++++-----
 kernel/bpf/verifier.c                              | 10 ++---
 kernel/trace/bpf_trace.c                           | 12 +++---
 kernel/trace/ftrace.c                              |  2 +
 tools/testing/selftests/bpf/prog_tests/recursion.c |  7 +++-
 tools/testing/selftests/bpf/test_progs.h           | 19 +++++++++
 9 files changed, 107 insertions(+), 48 deletions(-)
Signature · 1 line
1.8.3.1