[PATCH 1/2] iscsi: handle check condition status in retry loop

[PATCH 1/2] iscsi: handle check condition status in retry loop

[Xie Yongji]

The handling of check condition was incorrect because
we would only do it after retries exceed maximum.

Fixes: 8c460269aa ("iscsi: base all handling of check condition on scsi_sense_to_errno")
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
---
 block/iscsi.c | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/block/iscsi.c b/block/iscsi.c
index a8b76979d8..2964c9f8d2 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -266,16 +266,16 @@ iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
                 timer_mod(&iTask->retry_timer,
                           qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
                 iTask->do_retry = 1;
-            }
-        } else if (status == SCSI_STATUS_CHECK_CONDITION) {
-            int error = iscsi_translate_sense(&task->sense);
-            if (error == EAGAIN) {
-                error_report("iSCSI CheckCondition: %s",
-                             iscsi_get_error(iscsi));
-                iTask->do_retry = 1;
-            } else {
-                iTask->err_code = -error;
-                iTask->err_str = g_strdup(iscsi_get_error(iscsi));
+            } else if (status == SCSI_STATUS_CHECK_CONDITION) {
+                int error = iscsi_translate_sense(&task->sense);
+                if (error == EAGAIN) {
+                    error_report("iSCSI CheckCondition: %s",
+                                 iscsi_get_error(iscsi));
+                    iTask->do_retry = 1;
+                } else {
+                    iTask->err_code = -error;
+                    iTask->err_str = g_strdup(iscsi_get_error(iscsi));
+                }
             }
         }
     }
-- 
2.11.0

[PATCH 2/2] iscsi: return -EIO when sense fields are meaningless

[Xie Yongji]

When an I/O request failed, now we only return correct
value on scsi check condition. We should also have a
default errno such as -EIO in other case.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
---
 block/iscsi.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/block/iscsi.c b/block/iscsi.c
index 2964c9f8d2..387ed872ef 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -241,9 +241,11 @@ iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
 
     iTask->status = status;
     iTask->do_retry = 0;
+    iTask->err_code = 0;
     iTask->task = task;
 
     if (status != SCSI_STATUS_GOOD) {
+        iTask->err_code = -EIO;
         if (iTask->retries++ < ISCSI_CMD_RETRIES) {
             if (status == SCSI_STATUS_BUSY ||
                 status == SCSI_STATUS_TIMEOUT ||
-- 
2.11.0

Re: [PATCH 2/2] iscsi: return -EIO when sense fields are meaningless

[Paolo Bonzini]

On 01/07/20 12:54, Xie Yongji wrote:
> When an I/O request failed, now we only return correct
> value on scsi check condition. We should also have a
> default errno such as -EIO in other case.
> 
> Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
> ---
>  block/iscsi.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/block/iscsi.c b/block/iscsi.c
> index 2964c9f8d2..387ed872ef 100644
> --- a/block/iscsi.c
> +++ b/block/iscsi.c
> @@ -241,9 +241,11 @@ iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
>  
>      iTask->status = status;
>      iTask->do_retry = 0;
> +    iTask->err_code = 0;
>      iTask->task = task;
>  
>      if (status != SCSI_STATUS_GOOD) {
> +        iTask->err_code = -EIO;
>          if (iTask->retries++ < ISCSI_CMD_RETRIES) {
>              if (status == SCSI_STATUS_BUSY ||
>                  status == SCSI_STATUS_TIMEOUT ||
> 

Queued both, thanks.

Paolo