[RFC,bpf-next,v1,0/2] Fix deadlock with bpf_timer in fentry/fexit progs

Message

Kumar Kartikeya Dwivedi Nov. 6, 2022, 1:51 a.m. UTC

Hi, this is a fix for a deadlock condition I noticed (and verified by
constructing a test case, see patch #2) while looking at the code. There
might be alternative ways to solve this, like having percpu map ops
recursion counter, and detecting reentrancy based on that. I went for
the simpler option for now.

Previous discussions on a patch doing something similar for a different
issue here, but never added BPF_PROG_TYPE_TRACING to this list:

https://lore.kernel.org/bpf/CALOAHbBK1=[e-mail]
https://lore.kernel.org/bpf/CALOAHbAQ7AKNPrkawGKHarkB5CGkWXTC8P=+EJP3DAPNEf=[e-mail]

Note that timer and timer_crash selftests are broken by this change, so
it has potential for wider breakage.

Kumar Kartikeya Dwivedi (2):
  bpf: Fix deadlock for bpf_timer's spinlock
  [EXAMPLE] selftests/bpf: Add timer deadlock example

 kernel/bpf/verifier.c                         | 10 ++--
 .../testing/selftests/bpf/prog_tests/timer.c  | 21 +++++---
 .../selftests/bpf/prog_tests/timer_crash.c    |  9 +++-
 .../selftests/bpf/prog_tests/timer_deadlock.c | 29 +++++++++++
 tools/testing/selftests/bpf/progs/timer.c     |  8 +--
 .../testing/selftests/bpf/progs/timer_crash.c |  4 +-
 .../selftests/bpf/progs/timer_deadlock.c      | 50 +++++++++++++++++++
 7 files changed, 113 insertions(+), 18 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/timer_deadlock.c
 create mode 100644 tools/testing/selftests/bpf/progs/timer_deadlock.c


base-commit: 25906092edb4bcf94cb669bd1ed03a0ef2f4120c
--
2.38.1