On 21/08/2017 13:11, Fam Zheng wrote: > On Fri, 08/18 16:36, Paolo Bonzini wrote: >> On 18/08/2017 16:15, Fam Zheng wrote: >>> Two changes: >>> >>> 1) Look at asc/ascq for NOT_READY and DATA_PROTECT; >>> 2) Translate SPACE_ALLOC_FAILED as ENOSPC; >>> >>> Signed-off-by: Fam Zheng <[email protected]> >>> --- >>> util/scsi.c | 7 +++---- >>> 1 file changed, 3 insertions(+), 4 deletions(-) >>> >>> diff --git a/util/scsi.c b/util/scsi.c >>> index 92ca436cd0..d42ce33449 100644 >>> --- a/util/scsi.c >>> +++ b/util/scsi.c >>> @@ -18,13 +18,11 @@ >>> int scsi_sense_to_errno(int key, int asc, int ascq) >>> { >>> switch (key) { >>> - case 0x02: /* SCSI_SENSE_NOT_READY */ >>> - return EBUSY; >>> - case 0x07: /* SCSI_SENSE_DATA_PROTECTION */ >>> - return EACCES; >>> case 0x0b: /* SCSI_SENSE_COMMAND_ABORTED */ >>> return ECANCELED; >>> + case 0x02: /* SCSI_SENSE_NOT_READY */ >>> case 0x05: /* SCSI_SENSE_ILLEGAL_REQUEST */ >>> + case 0x07: /* SCSI_SENSE_DATA_PROTECTION */ >>> /* Parse ASCQ */ >>> break; >>> default: >> >> UNIT_ATTENTION should also be passed down to the guest without stopping >> the VM. Maybe map it to EAGAIN? > > Or just 0?
That works too, it depends on how you define 0. 1h is defined as informational only and can be dropped, 0x0401 or UNIT_ATTENTION definitely cannot. Paolo >> >> Looking at what Linux does: >> >> - RECOVERED_ERROR should just return 0 >> >> - 0x0401 is "unit in the process of becoming ready", and it should also >> be EAGAIN >> >> - 0x0402 is "initializing command required", and probably can be fixed >> by the guest by scsi-block but not by iscsi so it should be its own >> errno, maybe ENOTCONN? (iscsi might try sending START STOP UNIT >> followed by a bunch of TEST UNIT READYs, see scsi_eh_try_stu in Linux's >> drivers/scsi/scsi_error.c). > > These makes sense to me. > > Fam >
