Re: [PATCH lib] lib: alloc_tag_module_unload must wait for pending kfree_rcu calls

[Andrew Morton <akpm@linux-foundation.org>]

On Mon,  7 Oct 2024 22:52:24 +0200 Florian Westphal <fw@strlen.de> wrote:

> Ben Greear reports following splat:
>  ------------[ cut here ]------------
>  net/netfilter/nf_nat_core.c:1114 module nf_nat func:nf_nat_register_fn has 256 allocated at module unload
>  WARNING: CPU: 1 PID: 10421 at lib/alloc_tag.c:168 alloc_tag_module_unload+0x22b/0x3f0
>  Modules linked in: nf_nat(-) btrfs ufs qnx4 hfsplus hfs minix vfat msdos fat
> ...
>  Hardware name: Default string Default string/SKYBAY, BIOS 5.12 08/04/2020
>  RIP: 0010:alloc_tag_module_unload+0x22b/0x3f0
>   codetag_unload_module+0x19b/0x2a0
>   ? codetag_load_module+0x80/0x80
> 
> nf_nat module exit calls kfree_rcu on those addresses, but the free
> operation is likely still pending by the time alloc_tag checks for leaks.
> 
> Wait for outstanding kfree_rcu operations to complete before checking
> resolves this warning.
> 
> Reproducer:
> unshare -n iptables-nft -t nat -A PREROUTING -p tcp
> grep nf_nat /proc/allocinfo # will list 4 allocations
> rmmod nft_chain_nat
> rmmod nf_nat                # will WARN.
> 
> ...
>
> --- a/lib/codetag.c
> +++ b/lib/codetag.c
> @@ -228,6 +228,8 @@ bool codetag_unload_module(struct module *mod)
>  	if (!mod)
>  		return true;
>  
> +	kvfree_rcu_barrier();
> +
>  	mutex_lock(&codetag_lock);
>  	list_for_each_entry(cttype, &codetag_types, link) {
>  		struct codetag_module *found = NULL;

It's always hard to determine why a thing like this is present, so a
comment is helpful:

--- a/lib/codetag.c~lib-alloc_tag_module_unload-must-wait-for-pending-kfree_rcu-calls-fix
+++ a/lib/codetag.c
@@ -228,6 +228,7 @@ bool codetag_unload_module(struct module
 	if (!mod)
 		return true;
 
+	/* await any module's kfree_rcu() operations to complete */
 	kvfree_rcu_barrier();
 
 	mutex_lock(&codetag_lock);
_

But I do wonder whether this is in the correct place.

Waiting for a module's ->exit() function's kfree_rcu()s to complete
should properly be done by the core module handling code?

free_module() does a full-on synchronize_rcu() prior to freeing the
module memory itself and I think codetag_unload_module() could be
called after that?