Re: memory leaks of ftrace_hash

[Jiri Olsa <olsajiri@gmail.com>]

On Fri, Jan 30, 2026 at 07:50:13PM -0800, Alexei Starovoitov wrote:
> Hi Jiri,
> 
> I see a lot of:
> unreferenced object 0xff1100012c967540 (size 32):
>   comm "test_progs", pid 2052, jiffies 4294681345
>   hex dump (first 32 bytes):
>     00 00 00 00 00 00 00 00 b8 be 65 06 01 00 11 ff  ..........e.....
>     74 3f 5b 83 ff ff ff ff 80 0d 01 a0 ff ff ff ff  t?[.............
>   backtrace (crc 8dddf0a9):
>     __kmalloc_cache_noprof+0x3a5/0x520
>     add_ftrace_hash_entry_direct+0x4a/0x2e0
>     alloc_and_copy_ftrace_hash+0x15a/0x280
>     hash_sub+0x3a/0x470
>     update_ftrace_direct_del+0x349/0x7d0
>     bpf_trampoline_update+0xa9c/0x1070
>     bpf_trampoline_unlink_prog+0x287/0x600
>     bpf_tracing_link_release+0x5f/0x100
>     bpf_link_free+0xca/0x300
>     bpf_link_release+0x48/0x60
>     __fput+0x2fc/0x9d0
>     fput_close_sync+0xcd/0x170
>     __x64_sys_close+0x7d/0xd0
>     do_syscall_64+0x46/0xfc0
>     entry_SYSCALL_64_after_hwframe+0x4b/0x53
> 
> just running the latest bpf-next.
> 
> I suspect it's due to your recent changes in the area.
> 
> Pls take a look.

hi,
argh.. I had kmemleak disabled due to small CONFIG_DEBUG_KMEMLEAK_MEM_POOL_SIZE
looks like change below fixes it, I'll send a patch

thanks,
jirka


---
diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index 8574932e66b6..b12dbd93ae1c 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -6537,6 +6537,7 @@ int update_ftrace_direct_del(struct ftrace_ops *ops, struct ftrace_hash *hash)
 		/* free the new_direct_functions */
 		old_direct_functions = new_direct_functions;
 	} else {
+		old_direct_functions = direct_functions;
 		rcu_assign_pointer(direct_functions, new_direct_functions);
 	}