Re: inconsistent lock state in io_file_data_ref_zero

[Jens Axboe]

On 11/23/20 2:55 AM, syzbot wrote:
> Hello,
> 
> syzbot found the following issue on:
> 
> HEAD commit:    27bba9c5 Merge tag 'scsi-fixes' of git://git.kernel.org/pu..
> git tree:       upstream
> console output: https://syzkaller.appspot.com/x/log.txt?x=11041f1e500000
> kernel config:  https://syzkaller.appspot.com/x/.config?x=330f3436df12fd44
> dashboard link: https://syzkaller.appspot.com/bug?extid=1f4ba1e5520762c523c6
> compiler:       gcc (GCC) 10.1.0-syz 20200507
> syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=17d9b775500000
> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=157e4f75500000
> 
> The issue was bisected to:
> 
> commit dcd479e10a0510522a5d88b29b8f79ea3467d501
> Author: Johannes Berg <johannes.b...@intel.com>
> Date:   Fri Oct 9 12:17:11 2020 +0000
> 
>     mac80211: always wind down STA state

Not sure what is going on with the syzbot bisects recently, they are way
off into the weeds...

Anyway, I think the below should fix it.


diff --git a/fs/io_uring.c b/fs/io_uring.c
index 489ec7272b3e..0f2abbff7eec 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -7194,9 +7181,9 @@ static int io_sqe_files_unregister(struct io_ring_ctx 
*ctx)
        if (!data)
                return -ENXIO;
 
-       spin_lock(&data->lock);
+       spin_lock_bh(&data->lock);
        ref_node = data->node;
-       spin_unlock(&data->lock);
+       spin_unlock_bh(&data->lock);
        if (ref_node)
                percpu_ref_kill(&ref_node->refs);
 
@@ -7578,7 +7565,7 @@ static void io_file_data_ref_zero(struct percpu_ref *ref)
        data = ref_node->file_data;
        ctx = data->ctx;
 
-       spin_lock(&data->lock);
+       spin_lock_bh(&data->lock);
        ref_node->done = true;
 
        while (!list_empty(&data->ref_list)) {
@@ -7590,7 +7577,7 @@ static void io_file_data_ref_zero(struct percpu_ref *ref)
                list_del(&ref_node->node);
                first_add |= llist_add(&ref_node->llist, &ctx->file_put_llist);
        }
-       spin_unlock(&data->lock);
+       spin_unlock_bh(&data->lock);
 
        if (percpu_ref_is_dying(&data->refs))
                delay = 0;
@@ -7713,9 +7700,9 @@ static int io_sqe_files_register(struct io_ring_ctx *ctx, 
void __user *arg,
        }
 
        file_data->node = ref_node;
-       spin_lock(&file_data->lock);
+       spin_lock_bh(&file_data->lock);
        list_add_tail(&ref_node->node, &file_data->ref_list);
-       spin_unlock(&file_data->lock);
+       spin_unlock_bh(&file_data->lock);
        percpu_ref_get(&file_data->refs);
        return ret;
 out_fput:
@@ -7872,10 +7859,10 @@ static int __io_sqe_files_update(struct io_ring_ctx 
*ctx,
 
        if (needs_switch) {
                percpu_ref_kill(&data->node->refs);
-               spin_lock(&data->lock);
+               spin_lock_bh(&data->lock);
                list_add_tail(&ref_node->node, &data->ref_list);
                data->node = ref_node;
-               spin_unlock(&data->lock);
+               spin_unlock_bh(&data->lock);
                percpu_ref_get(&ctx->file_data->refs);
        } else
                destroy_fixed_file_ref_node(ref_node);


-- 
Jens Axboe