libiscsi: Fix host busy blocking during connection teardown
authorJohn Soni Jose <sony.john@avagotech.com>
Wed, 24 Jun 2015 01:11:58 +0000 (06:41 +0530)
committerJames Bottomley <JBottomley@Odin.com>
Wed, 12 Aug 2015 17:21:20 +0000 (10:21 -0700)
In case of hw iscsi offload, an host can have N-number of active
connections. There can be IO's running on some connections which
make host->host_busy always TRUE. Now if logout from a connection
is tried then the code gets into an infinite loop as host->host_busy
is always TRUE.

 iscsi_conn_teardown(....)
 {
   .........
    /*
     * Block until all in-progress commands for this connection
     * time out or fail.
     */
     for (;;) {
      spin_lock_irqsave(session->host->host_lock, flags);
      if (!atomic_read(&session->host->host_busy)) { /* OK for ERL == 0 */
      spin_unlock_irqrestore(session->host->host_lock, flags);
              break;
      }
     spin_unlock_irqrestore(session->host->host_lock, flags);
     msleep_interruptible(500);
     iscsi_conn_printk(KERN_INFO, conn, "iscsi conn_destroy(): "
                 "host_busy %d host_failed %d\n",
          atomic_read(&session->host->host_busy),
          session->host->host_failed);

................
...............
     }
  }

This is not an issue with software-iscsi/iser as each cxn is a separate
host.

Fix:
Acquiring eh_mutex in iscsi_conn_teardown() before setting
session->state = ISCSI_STATE_TERMINATE.

Signed-off-by: John Soni Jose <sony.john@avagotech.com>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Chris Leech <cleech@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: James Bottomley <JBottomley@Odin.com>
drivers/scsi/libiscsi.c

index 8053f24..98d9bb6 100644 (file)
@@ -2941,10 +2941,10 @@ void iscsi_conn_teardown(struct iscsi_cls_conn *cls_conn)
 {
        struct iscsi_conn *conn = cls_conn->dd_data;
        struct iscsi_session *session = conn->session;
-       unsigned long flags;
 
        del_timer_sync(&conn->transport_timer);
 
+       mutex_lock(&session->eh_mutex);
        spin_lock_bh(&session->frwd_lock);
        conn->c_stage = ISCSI_CONN_CLEANUP_WAIT;
        if (session->leadconn == conn) {
@@ -2956,28 +2956,6 @@ void iscsi_conn_teardown(struct iscsi_cls_conn *cls_conn)
        }
        spin_unlock_bh(&session->frwd_lock);
 
-       /*
-        * Block until all in-progress commands for this connection
-        * time out or fail.
-        */
-       for (;;) {
-               spin_lock_irqsave(session->host->host_lock, flags);
-               if (!atomic_read(&session->host->host_busy)) { /* OK for ERL == 0 */
-                       spin_unlock_irqrestore(session->host->host_lock, flags);
-                       break;
-               }
-               spin_unlock_irqrestore(session->host->host_lock, flags);
-               msleep_interruptible(500);
-               iscsi_conn_printk(KERN_INFO, conn, "iscsi conn_destroy(): "
-                                 "host_busy %d host_failed %d\n",
-                                 atomic_read(&session->host->host_busy),
-                                 session->host->host_failed);
-               /*
-                * force eh_abort() to unblock
-                */
-               wake_up(&conn->ehwait);
-       }
-
        /* flush queued up work because we free the connection below */
        iscsi_suspend_tx(conn);
 
@@ -2994,6 +2972,7 @@ void iscsi_conn_teardown(struct iscsi_cls_conn *cls_conn)
        if (session->leadconn == conn)
                session->leadconn = NULL;
        spin_unlock_bh(&session->frwd_lock);
+       mutex_unlock(&session->eh_mutex);
 
        iscsi_destroy_conn(cls_conn);
 }