> On Oct 24, 2025, at 4:43 AM, Jiri Olsa <[email protected]> wrote:
>
> On Fri, Oct 24, 2025 at 12:12:56AM -0700, Song Liu wrote:
>> ftrace_hash_ipmodify_enable() checks IPMODIFY and DIRECT ftrace_ops on
>> the same kernel function. When needed, ftrace_hash_ipmodify_enable()
>> calls ops->ops_func() to prepare the direct ftrace (BPF trampoline) to
>> share the same function as the IPMODIFY ftrace (livepatch).
>>
>> ftrace_hash_ipmodify_enable() is called in register_ftrace_direct() path,
>> but not called in modify_ftrace_direct() path. As a result, the following
>> operations will break livepatch:
>>
>> 1. Load livepatch to a kernel function;
>> 2. Attach fentry program to the kernel function;
>> 3. Attach fexit program to the kernel function.
>>
>> After 3, the kernel function being used will not be the livepatched
>> version, but the original version.
>>
>> Fix this by adding ftrace_hash_ipmodify_enable() to modify_ftrace_direct()
>> and adjust some logic around the call.
>>
>> Signed-off-by: Song Liu <[email protected]>
>> ---
>> kernel/bpf/trampoline.c | 12 +++++++-----
>> kernel/trace/ftrace.c | 12 ++++++++++--
>> 2 files changed, 17 insertions(+), 7 deletions(-)
>>
>> diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c
>> index 5949095e51c3..8015f5dc3169 100644
>> --- a/kernel/bpf/trampoline.c
>> +++ b/kernel/bpf/trampoline.c
>> @@ -221,6 +221,13 @@ static int register_fentry(struct bpf_trampoline *tr,
>> void *new_addr)
>>
>> if (tr->func.ftrace_managed) {
>> ftrace_set_filter_ip(tr->fops, (unsigned long)ip, 0, 1);
>> + /*
>> + * Clearing fops->trampoline_mutex and fops->NULL is
>
> s/trampoline_mutex/trampoline/
Good catch!
>
>> + * needed by the "goto again" case in
>> + * bpf_trampoline_update().
>> + */
>> + tr->fops->trampoline = 0;
>> + tr->fops->func = NULL;
>
> IIUC you move this because if modify_fentry returns -EAGAIN
> we don't want to reset the trampoline, right?
Right, we don’t want to reset this in the modify_fentry path.
We can add a check before “goto again” so that we only do the
reset for register_fentry, but I think it is cleaner this way.
I can be convinced to change it.
>
>> ret = register_ftrace_direct(tr->fops, (long)new_addr);
>> } else {
>> ret = bpf_arch_text_poke(ip, BPF_MOD_CALL, NULL, new_addr);
>> @@ -479,11 +486,6 @@ static int bpf_trampoline_update(struct bpf_trampoline
>> *tr, bool lock_direct_mut
>> * BPF_TRAMP_F_SHARE_IPMODIFY is set, we can generate the
>> * trampoline again, and retry register.
>> */
>> - /* reset fops->func and fops->trampoline for re-register */
>> - tr->fops->func = NULL;
>> - tr->fops->trampoline = 0;
>> -
>> - /* free im memory and reallocate later */
>> bpf_tramp_image_free(im);
>> goto again;
>> }
>> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
>> index 7f432775a6b5..370f620734cf 100644
>> --- a/kernel/trace/ftrace.c
>> +++ b/kernel/trace/ftrace.c
>> @@ -2020,8 +2020,6 @@ static int __ftrace_hash_update_ipmodify(struct
>> ftrace_ops *ops,
>> if (is_ipmodify)
>> goto rollback;
>>
>> - FTRACE_WARN_ON(rec->flags & FTRACE_FL_DIRECT);
>
> why is this needed?
This is needed for the modify_ftrace_direct case, because
the record already have a direct function (BPF trampoline)
attached.
Thanks,
Song
