On 8/9/21 10:18 AM, Andrii Nakryiko wrote:
On Sun, Aug 8, 2021 at 11:03 PM Yonghong Song <yhs@xxxxxx> wrote:Currently, if bpf_get_current_cgroup_id() or bpf_get_current_ancestor_cgroup_id() helper is called with sleepable programs e.g., sleepable fentry/fmod_ret/fexit/lsm programs, a rcu warning may appear. For example, if I added the following hack to test_progs/test_lsm sleepable fentry program test_sys_setdomainname: --- a/tools/testing/selftests/bpf/progs/lsm.c +++ b/tools/testing/selftests/bpf/progs/lsm.c @@ -168,6 +168,10 @@ int BPF_PROG(test_sys_setdomainname, struct pt_regs *regs) int buf = 0; long ret; + __u64 cg_id = bpf_get_current_cgroup_id(); + if (cg_id == 1000) + copy_test++; + ret = bpf_copy_from_user(&buf, sizeof(buf), ptr); if (len == -2 && ret == 0 && buf == 1234) copy_test++; I will hit the following rcu warning: include/linux/cgroup.h:481 suspicious rcu_dereference_check() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 1 lock held by test_progs/260: #0: ffffffffa5173360 (rcu_read_lock_trace){....}-{0:0}, at: __bpf_prog_enter_sleepable+0x0/0xa0 stack backtrace: CPU: 1 PID: 260 Comm: test_progs Tainted: G O 5.14.0-rc2+ #176 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a1990b-prebuilt.qemu.org 04/01/2014 Call Trace: dump_stack_lvl+0x56/0x7b bpf_get_current_cgroup_id+0x9c/0xb1 bpf_prog_a29888d1c6706e09_test_sys_setdomainname+0x3e/0x89c bpf_trampoline_6442469132_0+0x2d/0x1000 __x64_sys_setdomainname+0x5/0x110 do_syscall_64+0x3a/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae I can get similar warning using bpf_get_current_ancestor_cgroup_id() helper. syzbot reported a similar issue in [1] for syscall program. Helper bpf_get_current_cgroup_id() or bpf_get_current_ancestor_cgroup_id() has the following callchain: task_dfl_cgroup task_css_set task_css_set_check and we have #define task_css_set_check(task, __c) \ rcu_dereference_check((task)->cgroups, \ lockdep_is_held(&cgroup_mutex) || \ lockdep_is_held(&css_set_lock) || \ ((task)->flags & PF_EXITING) || (__c)) Since cgroup_mutex/css_set_lock is not held and the task is not existing and rcu read_lock is not held, a warning will be issued. Note that bpf sleepable program is protected by rcu_read_lock_trace(). To fix the issue, let us make these two helpers not available to sleepable program. I marked the patch fixing 95b861a7935b ("bpf: Allow bpf_get_current_ancestor_cgroup_id for tracing") which added bpf_get_current_ancestor_cgroup_id() to 5.14. I think backporting 5.14 is probably good enough as sleepable progrems are not widely used. This patch should fix [1] as well since syscall program is a sleepable program and bpf_get_current_cgroup_id() is not available to syscall program any more. [1] https://lore.kernel.org/bpf/0000000000006d5cab05c7d9bb87@xxxxxxxxxx/ Reported-by: syzbot+7ee5c2c09c284495371f@xxxxxxxxxxxxxxxxxxxxxxxxx Fixes: 95b861a7935b ("bpf: Allow bpf_get_current_ancestor_cgroup_id for tracing") Signed-off-by: Yonghong Song <yhs@xxxxxx> --- kernel/trace/bpf_trace.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c index b4916ef388ad..eaa8a8ffbe46 100644 --- a/kernel/trace/bpf_trace.c +++ b/kernel/trace/bpf_trace.c @@ -1016,9 +1016,11 @@ bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) #endif #ifdef CONFIG_CGROUPS case BPF_FUNC_get_current_cgroup_id: - return &bpf_get_current_cgroup_id_proto; + return prog->aux->sleepable ? + NULL : &bpf_get_current_cgroup_id_proto; case BPF_FUNC_get_current_ancestor_cgroup_id: - return &bpf_get_current_ancestor_cgroup_id_proto; + return prog->aux->sleepable ? + NULL : &bpf_get_current_ancestor_cgroup_id_proto;This feels too extreme. I bet these helpers are as useful in sleepable BPF progs as they are in non-sleepable ones. Why don't we just implement a variant of get_current_cgroup_id (and the ancestor variant as well) which takes that cgroup_mutex lock, and just pick the appropriate implementation. Wouldn't that work?
This may not work. e.g., for sleepable fentry program, if the to-be-traced function is inside in cgroup_mutex, we will have a deadlock. Currently, affected program types are tracing/fentry.s, tracing/fexit.s, tracing/fmod_ret.s, lsm.s and syscall. For fmod_ret.s, lsm.s, they all have some kind of predefined attachment/context, we might be able to check all potential attachment points and allow these two helpers when attachment point is not surrounded by cgroup_mutex. For syscall program, we should be okay as it is called with bpf_prog_test_run interface but I am not sure why user wants a cgroup_id for that.
#endif case BPF_FUNC_send_signal: return &bpf_send_signal_proto; -- 2.30.2
