Re: sound: list corruption in delete_and_unsubscribe_port

[Takashi Iwai]

On Tue, 16 Feb 2016 10:41:54 +0100,
Dmitry Vyukov wrote:
> 
> Hello,
> 
> Here is a new one on 18558cae0272f8fd9647e69d3fec1565a7949865
> (4.5-rc4). But need to note that sound become much more stable, I've
> seen only 2 of these over night.
> 
> The following program causes list corruption:
> 
> ------------[ cut here ]------------
> WARNING: CPU: 2 PID: 12546 at lib/list_debug.c:62 __list_del_entry+0x10b/0x1e0()
> list_del corruption, ffff880063512388->next is LIST_POISON1 (dead000000000100)
> Modules linked in:
> CPU: 2 PID: 12546 Comm: a.out Not tainted 4.5.0-rc4+ #328
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
>  ffffffff87b05080 ffff8800608b7a48 ffffffff82be46cf ffffffff81477fb8
>  fffffbfff0f60a10 ffff8800608b7ab8 ffff8800637d97c0 ffffffff86ad3780
>  0000000000000009 000000000000003e ffff8800608b7a88 ffffffff81355139
> Call Trace:
>  [<ffffffff81355249>] warn_slowpath_fmt+0xa9/0xd0 kernel/panic.c:494
>  [<ffffffff82c4c36b>] __list_del_entry+0x10b/0x1e0 lib/list_debug.c:60
>  [<ffffffff82c4c44d>] list_del+0xd/0x70 lib/list_debug.c:86
>  [<ffffffff852c38e3>] delete_and_unsubscribe_port+0x1e3/0x2f0
> sound/core/seq/seq_ports.c:545
>  [<ffffffff852c43fa>] clear_subscriber_list+0x15a/0x260
> sound/core/seq/seq_ports.c:250
>  [<ffffffff852c456a>] port_delete+0x6a/0x1c0 sound/core/seq/seq_ports.c:266
>  [<ffffffff852c5242>] snd_seq_delete_all_ports+0x242/0x350
> sound/core/seq/seq_ports.c:330
>  [<ffffffff852ae1cf>] seq_free_client1+0x2f/0x290
> sound/core/seq/seq_clientmgr.c:272
>  [<ffffffff852ae495>] seq_free_client+0x65/0x160
> sound/core/seq/seq_clientmgr.c:299
>  [<ffffffff852b118d>] snd_seq_release+0x4d/0xb0
> sound/core/seq/seq_clientmgr.c:380
>  [<ffffffff817c3256>] __fput+0x236/0x780 fs/file_table.c:208
>  [<ffffffff817c3825>] ____fput+0x15/0x20 fs/file_table.c:244
>  [<ffffffff813b3100>] task_work_run+0x170/0x210 kernel/task_work.c:115
>  [<     inline     >] tracehook_notify_resume include/linux/tracehook.h:191
>  [<ffffffff810066b1>] exit_to_usermode_loop+0x1d1/0x210
> arch/x86/entry/common.c:251
>  [<     inline     >] prepare_exit_to_usermode arch/x86/entry/common.c:282
>  [<ffffffff810084ea>] syscall_return_slowpath+0x2ba/0x340
> arch/x86/entry/common.c:344
>  [<ffffffff866626e2>] int_ret_from_sys_call+0x25/0x9f
> arch/x86/entry/entry_64.S:281
> ---[ end trace 4cad985f706f8ace ]---

Hm, this might be the remaining open race at deleting ports.
Please try the patch below.


thanks,

Takashi

---
diff --git a/sound/core/seq/seq_ports.c b/sound/core/seq/seq_ports.c
index 921fb2bd8fad..fe686ee41c6d 100644
--- a/sound/core/seq/seq_ports.c
+++ b/sound/core/seq/seq_ports.c
@@ -535,19 +535,22 @@ static void delete_and_unsubscribe_port(struct snd_seq_client *client,
 					bool is_src, bool ack)
 {
 	struct snd_seq_port_subs_info *grp;
+	struct list_head *list;
+	bool empty;
 
 	grp = is_src ? &port->c_src : &port->c_dest;
+	list = is_src ? &subs->src_list : &subs->dest_list;
 	down_write(&grp->list_mutex);
 	write_lock_irq(&grp->list_lock);
-	if (is_src)
-		list_del(&subs->src_list);
-	else
-		list_del(&subs->dest_list);
+	empty = list_empty(list);
+	if (!empty)
+		list_del_init(list);
 	grp->exclusive = 0;
 	write_unlock_irq(&grp->list_lock);
 	up_write(&grp->list_mutex);
 
-	unsubscribe_port(client, port, grp, &subs->info, ack);
+	if (!empty)
+		unsubscribe_port(client, port, grp, &subs->info, ack);
 }
 
 /* connect two ports */