Re: [PATCH] uprobes: get RCU trace lock before list iteration

[Peter Zijlstra]

On Thu, Nov 07, 2024 at 09:14:45AM -0800, Breno Leitao wrote:
> Acquire RCU trace lock in filter_chain() to protect
> list_for_each_entry_rcu() iteration, protecting the list iteration in a
> RCU read section.
> 
> Prior to this fix, list_for_each_entry_srcu() was called without holding
> the required lock, triggering warnings when RCU_PROVING is enabled:
> 
>       kernel/events/uprobes.c:937 RCU-list traversed without holding the 
> required lock!!
> 
> Signed-off-by: Breno Leitao <lei...@debian.org>
> Fixes: cc01bd044e6a ("uprobes: travers uprobe's consumer list locklessly 
> under SRCU protection")
> ---
>  kernel/events/uprobes.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> index 
> fa04b14a7d72353adc440742016b813da6c812d2..afdaa45a43ac3948f7983175eda808c989e8738a
>  100644
> --- a/kernel/events/uprobes.c
> +++ b/kernel/events/uprobes.c
> @@ -1103,11 +1103,13 @@ static bool filter_chain(struct uprobe *uprobe, 
> struct mm_struct *mm)
>       bool ret = false;
>  
>       down_read(&uprobe->consumer_rwsem);
> +     rcu_read_lock_trace();
>       list_for_each_entry_rcu(uc, &uprobe->consumers, cons_node, 
> rcu_read_lock_trace_held()) {

Maybe I'm confused, but isn't uprobe->consumer list protected by
uprobe->consumer_rwsem, which we hold for reading?

That is, AFAICT this is a false positive and we should be doing this
instead, no?


diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index a76ddc5fc982..a5405e9ef9f5 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -1104,7 +1104,7 @@ static bool filter_chain(struct uprobe *uprobe, struct 
mm_struct *mm)
        bool ret = false;
 
        down_read(&uprobe->consumer_rwsem);
-       list_for_each_entry_rcu(uc, &uprobe->consumers, cons_node, 
rcu_read_lock_trace_held()) {
+       list_for_each_entry(uc, &uprobe->consumers, cons_node) {
                ret = consumer_filter(uc, mm);
                if (ret)
                        break;