Instead of calling into scsi_handle_rw_error() directly from
scsi_block_sgio_complete() and skipping the normal callback, go through
the normal cleanup path by calling the callback with a positive error
value.

The important difference here is not only that the code path is cleaner,
but that the callbacks set r->req.aiocb = NULL. If we skip setting this
and the error action is BLOCK_ERROR_ACTION_STOP, resuming the VM runs
into an assertion failure in scsi_read_data() or scsi_write_data()
because the dangling aiocb pointer is unexpected.

Fixes: a108557bbf ("scsi: inline sg_io_sense_from_errno() into the callers.")
Buglink: https://issues.redhat.com/browse/RHEL-50000
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Message-ID: <20240731123207.27636-3-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
 hw/scsi/scsi-disk.c | 10 ----------
 1 file changed, 10 deletions(-)

diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
index 3ff6798bde..6e1a5c98df 100644
--- a/hw/scsi/scsi-disk.c
+++ b/hw/scsi/scsi-disk.c
@@ -2832,16 +2832,6 @@ static void scsi_block_sgio_complete(void *opaque, int ret)
         } else {
             ret = io_hdr->status;
         }
-
-        if (ret > 0) {
-            if (scsi_handle_rw_error(r, ret, true)) {
-                scsi_req_unref(&r->req);
-                return;
-            }
-
-            /* Ignore error.  */
-            ret = 0;
-        }
     }
 
     req->cb(req->cb_opaque, ret);
-- 
2.45.2