[Drbd-dev] [BUG] block: drbd: four possible ABBA deadlocks

Jia-Ju Bai baijiaju1990 at gmail.com
Thu Aug 19 11:31:04 CEST 2021


Hello,

My static analysis tool reports four possible ABBA deadlocks in the drbd 
drivers in Linux 5.10:

### Deadlock A ###
w_after_conn_state_ch()
   mutex_lock(&connection->resource->conf_update); --> line 2065
   conn_free_crypto()
     drbd_free_sock()
       drbd_free_one_sock(&connection->data)
         mutex_lock(&ds->mutex); --> line 2967

drbd_adm_net_opts()
   mutex_lock(&connection->data.mutex); --> line 2445
   mutex_lock(&connection->resource->conf_update); --> line 2446

When w_after_conn_state_ch() and drbd_adm_net_opts() are concurrently 
executed, the deadlock can occur.

### Deadlock B ###
drbd_adm_connect()
   mutex_lock(&adm_ctx.resource->conf_update); --> line 2644
   conn_free_crypto()
     drbd_free_one_sock(&connection->data)
       mutex_lock(&ds->mutex); --> line 2967

drbd_adm_net_opts()
   mutex_lock(&connection->data.mutex); --> line 2445
   mutex_lock(&connection->resource->conf_update); --> line 2446

When drbd_adm_connect() and drbd_adm_net_opts() are concurrently 
executed, the deadlock can occur.

### Deadlock C ###
w_after_conn_state_ch()
   mutex_lock(&connection->resource->conf_update); --> line 2065
   conn_free_crypto()
     drbd_free_sock()
       drbd_free_one_sock(&connection->data)
         mutex_lock(&ds->mutex); --> line 2967

receive_protocol()
   mutex_lock(&connection->data.mutex); --> line 3778
   mutex_lock(&connection->resource->conf_update); --> line 3779

When w_after_conn_state_ch() and receive_protocol() are concurrently 
executed, the deadlock can occur.

### Deadlock D ###
drbd_adm_connect()
   mutex_lock(&adm_ctx.resource->conf_update); --> line 2644
   conn_free_crypto()
     drbd_free_one_sock(&connection->data)
       mutex_lock(&ds->mutex); --> line 2967

receive_protocol()
   mutex_lock(&connection->data.mutex); --> line 3778
   mutex_lock(&connection->resource->conf_update); --> line 3779

When drbd_adm_connect() and receive_protocol() are concurrently 
executed, the deadlock can occur.

I am not quite sure whether these possible deadlocks are real and how to 
fix them if they are real.
Any feedback would be appreciated, thanks

Reported-by: TOTE Robot <oslab at tsinghua.edu.cn>


Best wishes,
Jia-Ju Bai


More information about the drbd-dev mailing list