Re: [PATCHv2 bpf-next 5/8] ftrace: Add update_ftrace_direct_mod function

[Jiri Olsa <olsajiri@gmail.com>]

On Thu, Nov 13, 2025 at 01:02:17PM +0000, bot+bpf-ci@kernel.org wrote:
> > diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h
> > index 433c36c3a..bacb6d9ab 100644
> > --- a/include/linux/ftrace.h
> > +++ b/include/linux/ftrace.h
> > @@ -544,6 +544,7 @@ int modify_ftrace_direct_nolock(struct ftrace_ops *ops, unsigned long addr);
> >
> >  int update_ftrace_direct_add(struct ftrace_ops *ops, struct ftrace_hash *hash);
> >  int update_ftrace_direct_del(struct ftrace_ops *ops, struct ftrace_hash *hash);
> > +int update_ftrace_direct_mod(struct ftrace_ops *ops, struct ftrace_hash *hash, bool do_direct_lock);
> >
> >  void ftrace_stub_direct_tramp(void);
> >
> > @@ -581,6 +582,11 @@ int update_ftrace_direct_del(struct ftrace_ops *ops, struct ftrace_hash *hash)
> >  	return -ENODEV;
> >  }
> >
> > +int modify_ftrace_direct_hash(struct ftrace_ops *ops, struct ftrace_hash *hash, bool do_direct_lock)
>                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
> The stub function name doesn't match the declaration. Should this be
> update_ftrace_direct_mod() instead of modify_ftrace_direct_hash()?

it should, my bad

> 
> > +{
> > +	return -ENODEV;
> > +}
> > +
> 
> [ ... ]
> 
> > diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
> > index 345f3fdd2..627d05b57 100644
> > --- a/kernel/trace/ftrace.c
> > +++ b/kernel/trace/ftrace.c
> > @@ -6445,6 +6445,74 @@ int update_ftrace_direct_del(struct ftrace_ops *ops, struct ftrace_hash *hash)
> >  	return err;
> >  }
> >
> > +int update_ftrace_direct_mod(struct ftrace_ops *ops, struct ftrace_hash *hash, bool do_direct_lock)
> > +{
> > +	struct ftrace_hash *orig_hash = ops->func_hash->filter_hash;
> > +	struct ftrace_func_entry *entry, *tmp;
> > +	static struct ftrace_ops tmp_ops = {
> > +		.func		= ftrace_stub,
> > +		.flags		= FTRACE_OPS_FL_STUB,
> > +	};
> > +	unsigned long size, i;
> > +	int err;
> > +
> > +	if (!hash_count(hash))
> > +		return 0;
> > +	if (check_direct_multi(ops))
> > +		return -EINVAL;
> > +	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
> > +		return -EINVAL;
> > +	if (direct_functions == EMPTY_HASH)
> > +		return -EINVAL;
> > +
> > +	if (do_direct_lock)
> > +		mutex_lock(&direct_mutex);
> > +
> > +	/* Enable the tmp_ops to have the same functions as the direct ops */
> > +	ftrace_ops_init(&tmp_ops);
> > +	tmp_ops.func_hash = ops->func_hash;
> > +
> > +	err = register_ftrace_function_nolock(&tmp_ops);
> > +	if (err)
> > +		goto unlock;
> > +
> > +	/*
> > +	 * Call __ftrace_hash_update_ipmodify() here, so that we can call
> > +	 * ops->ops_func for the ops. This is needed because the above
> > +	 * register_ftrace_function_nolock() worked on tmp_ops.
> > +	 */
> > +	err = __ftrace_hash_update_ipmodify(ops, orig_hash, orig_hash, true);
>                                                                         ^^^^
> Does __ftrace_hash_update_ipmodify() accept a fourth argument? The
> function signature shows it only takes three parameters (ops, old_hash,
> new_hash). This looks like a compilation error.

the whole patchset is based on bpf-next/master plus Song's livepatch
fixes which change modify_ftrace_direct_hash function, it's mentioned
in the cover letter

> 
> > +	if (err)
> > +		goto out;
> 
> [ ... ]
> 
> > +
> > +out:
> > +	/* Removing the tmp_ops will add the updated direct callers to the functions */
> > +	unregister_ftrace_function(&tmp_ops);
>          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> Can this deadlock when do_direct_lock is true? Looking at the code:
> 
>   update_ftrace_direct_mod() with do_direct_lock=true
>     -> mutex_lock(&direct_mutex)
>     -> register_ftrace_function_nolock(&tmp_ops) succeeds
>     -> __ftrace_hash_update_ipmodify() fails
>     -> goto out (still holding direct_mutex)
>     -> unregister_ftrace_function(&tmp_ops)
>        -> cleanup_direct_functions_after_ipmodify()
>           -> mutex_lock(&direct_mutex) /* deadlock */
> 
> The unregister path tries to reacquire direct_mutex that we're already
> holding.

I don't think this can happen because tmp_ops does not have
FTRACE_OPS_FL_IPMODIFY flag set

thanks,
jirka

> 
> > +
> > +unlock:
> > +	if (do_direct_lock)
> > +		mutex_unlock(&direct_mutex);
> > +	return err;
> > +}
> > +
> >  #endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
> 
> 
> ---
> AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
> See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md
> 
> CI run summary: https://github.com/kernel-patches/bpf/actions/runs/19332026793