[PATCH] SCSI: Prevent from retrying with expecting_cc_ua in case of disk change.

[PATCH] SCSI: Prevent from retrying with expecting_cc_ua in case of disk change.

[TARUISI Hiroaki]

Hi,

 I have found a following trouble.

   1. Boot with a particular usb DVD-RAM drive with RHEL boot disk
      in rescue mode.
   2. Change disk to DVD-RAM (whatever disk may ok which is bigger
      capacity than previous boot disk)
   3. execute ioctl(BLKGETSIZE[64]).
   4. We get the size of boot disk. (not one of DVD-RAM)

  Actually, I replaced to DVD-RAM disk with ext2 filesystem and check
  it with fsck, I get these nice error message.

      The filesystem size (according to the superblock) is xxxx blocks
      The physical size of the device is yyyy blocks
      Either the superblock or the partition table is likely to be corrupt!
      Abort?

  I found this problem is caused by retry logic of scsi driver.
  scsi driver retries i-o in case of unit attention with expecting_cc_ua=1
  even if the unit attention means disk change. So, sr driver loses unit
  attention, and also loses disk change event.

  (As for expecting_cc_ua=1, this was set because of a hardware bug and
   this is also another cause. I wonder why this bit is off only when
   unit attention or not ready. It should be off at the first i-o after
   reset operation, I think...)

  I made a simple patch for this problem. Could someone check this patch?

Regards,
taruisi

Signed-off-by: TARUISI Hiroaki <taruishi.hiroak@jp.fujitsu.com>
---
 drivers/scsi/scsi_error.c |   12 +++++++-----
 1 files changed, 7 insertions(+), 5 deletions(-)

diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 30ac116..fd93145 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -285,13 +285,15 @@ static int scsi_check_sense(struct scsi_cmnd *scmd)
    case UNIT_ATTENTION:
        /*
         * if we are expecting a cc/ua because of a bus reset that we
-        * performed, treat this just as a retry.  otherwise this is
-        * information that we should pass up to the upper-level driver
-        * so that we can deal with it there.
+        * performed, treat this just as a retry except a case of disk
+        * change.  otherwise this is information that we should pass up
+        * to the upper-level driver so that we can deal with it there.
         */
        if (scmd->device->expecting_cc_ua) {
-           scmd->device->expecting_cc_ua = 0;
-           return NEEDS_RETRY;
+           if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
+               scmd->device->expecting_cc_ua = 0;
+               return NEEDS_RETRY;
+           }
        }
        /*
         * if the device is in the process of becoming ready, we
-- 
1.4.4.4


-- 
taruisi

[PATCH] SCSI: Prevent from retrying with expecting_cc_ua in case of disk change.

[TARUISI Hiroaki]

Hi James,

Thanks for your comment.
Actually, Rob and I are focusing to the same problem.
And I have a dvd drive of that kind.

To confirm ASC,ASCQ in this case, I added some printk
like this.
----
        * so that we can deal with it there.
        */
       if (scmd->device->expecting_cc_ua) {
               printk("scsi_check_sense: expecting_cc_ua=on"
                      " unit_attention asc=%02x ascq=%02x\n",
                      sshdr.asc, sshdr.ascq);
               ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
               scmd->device->expecting_cc_ua = 0;
               return SCSI_MLQUEUE_DIS_DEV_RETRY;
       }
       /*
        * if the device is in the process of becoming ready, we
        * should retry.
        */
----
And I created boot dvd, booted the system with that in
rescue mode, changed the media after boot and read or write
that media, then, I could see these messages.
----
<6>Adding 4095992k swap on /dev/VolGroup00/LogVol01.  Priority:-1 extents:1 across:4095996k
<4>scsi_check_sense: expecting_cc_ua=on unit_attention asc=28 ascq=00
<4>end_request: I/O error, dev sr0, sector 7432832
----

As message says, my dvd drive reports media change sense and
as the result end_request I/O error is caused.
Surely this drive may be naughty, but we should know what kind of
UA and treat it.


---
 drivers/scsi/scsi_error.c |   11 +++++++++--
 1 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index a4b9cdb..158ba46 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -293,8 +293,15 @@ static int scsi_check_sense(struct scsi_cmnd *scmd)
                 * so that we can deal with it there.
                 */
                if (scmd->device->expecting_cc_ua) {
-                       scmd->device->expecting_cc_ua = 0;
-                       return NEEDS_RETRY;
+               /*
+                * Because some device does not queue unit attentions correctly,
+                * we carefully check additional sense code and qualifier
+                * so as not to squash media change unit attention.
+                */
+                       if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
+                               scmd->device->expecting_cc_ua = 0;
+                               return NEEDS_RETRY;
+                       }
                }
                /*
                 * if the device is in the process of becoming ready, we
--
1.6.5

(2011/08/10 0:22), James Bottomley wrote:
> On Tue, 2011-08-09 at 07:50 +0900, TARUISI Hiroaki wrote:
>> Hi James,
>>
>> I posted a patch (as below) last year about this.
>>   http://www.kerneltrap.org/mailarchive/linux-scsi/2010/12/22/6887895
>>
>> I intended to distinguish UA caused by media change from UA caused by
>> other reasons like resetting a device. First, it should be notified to upper
>> layer driver, sr, but the other must be retried as scsi_error.c coded.
>>
>> At least SCSI driver should not squash all UAs, I think.
> 
> Well ... the spec says the device should stack UAs in this event.
> However, I can totally believe that a broken CD would fail on this
> front.
> 
> If Mike and Rob can verify that it's realy asc/ascq 0x28/0x0 that the CD
> is emitting, I'd be happy to put the patch in (with a nice big comment
> explaining what's being done and why over the if() check rather than
> updating the global comment).
> 
> James
> 
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
> 
> 
-- 
taruisi