Hi Lance,
On Mon, 2014-11-24 at 12:40 +0000, Lance Gropper wrote:
> Hello LIO friends:
>
> A co-worker and myself are both trying LIO for the first time on a pair
> of systems running Ubuntu Server LTS 14.04.1. We are using QLogic Fibre
> targets (with iblock devices) and are having different problems, both of
> which only occur when a Windows-based initiator connects:
>
> On my co-workers' system, he is seeing repeating messages indicating:
>
> SCSI OP 28h with too big sectors 16384 exceeds fabric_max_sectors: 8192
>
So the MSFT initiator (Windows version..?) is sending individual I/Os
with a payload size of 16384 sectors * 512 bytes per sector = 8MB, and
hitting the hard-coded fabric_max_sectors limit of 4 MB in target-core.
There are two options for a work-around:
First option is to limit the max I/O size in the FC/SCSI miniport driver
on the MSFT initiator.
Depending on your version of Windows, this will involve a registry
change to reduce the current MaximumSGist or IIRC newer Windows versions
have a different way of doing this. You'll need google around based on
your FC HBA driver to figure out how to do this.
The second option is to bump the hard-coded fabric_max_sectors limit
within target-core. Here's a quick patch:
diff --git a/include/target/target_core_base.h b/include/target/target_core_base.h
index 23c518a..4445d97 100644
--- a/include/target/target_core_base.h
+++ b/include/target/target_core_base.h
@@ -78,7 +78,7 @@
/* Default max_write_same_len, disabled by default */
#define DA_MAX_WRITE_SAME_LEN 0
/* Default max transfer length */
-#define DA_FABRIC_MAX_SECTORS 8192
+#define DA_FABRIC_MAX_SECTORS 16384
/* Use a model alias based on the configfs backend device name */
#define DA_EMULATE_MODEL_ALIAS 0
/* Emulation for Direct Page Out */
@@ -111,7 +111,7 @@
/* Force SPC-3 PR Activate Persistence across Target Power Loss */
#define DA_FORCE_PR_APTPL 0
#define DA_STATUS_MAX_SECTORS_MIN 16
-#define DA_STATUS_MAX_SECTORS_MAX 8192
+#define DA_STATUS_MAX_SECTORS_MAX 16384
/* By default don't report non-rotating (solid state) medium */
#define DA_IS_NONROT 0
/* Queue Algorithm Modifier default for restricted reordering in control mode page */
So we've been limiting I/Os to 4 MB using 512 byte sectors in recent
times from a v3.4 kernel change introduced by Roland (CC'ed) in:
commit 015487b89f27d91d95a056cdc3c85e6c729bff12
Author: Roland Dreier <roland@xxxxxxxxxxxxxxx>
Date: Mon Feb 13 16:18:17 2012 -0800
target: Untangle front-end and back-end meanings of max_sectors attribute
However, most IBLOCK backends can use the block layer to split a 8 MB
payload into multiple I/Os as necessary outside of target-core.
FILEIO is also be able to handle up to 8 MB I/Os as well.
All that said, this issue will need to be addressed in mainline if new
versions of Windows are generating FC I/Os with 8 MB payloads using 512
byte sectors.
> On mine, I get different messages: When I add the target port, I get:
>
> **** Failed mbx[0]=4005, mb[1]=4, mb[2]=8800, mb[3]=ffff, cmd=70 ****.
> Failed to enable receiving of RSCN request: 0x2.
>
IIRC this RSCN failure is normal in Point-to-Point (eg: no FC switch)
configurations.
> Then when I try to connect with the Windows Initiator, I get:
>
> TARGET_CORE[qla2xxx]: Expected Transfer Length: 512 does not match SCSI
> CDB Length: 8 for SAM Opcode: 0x25
>
Warning about fabric level EDTL vs. CDB length mismatch for the incoming
READ_CAPACITY command. Can be safely ignored.
Thanks for reporting,
--nab
--
To unsubscribe from this list: send the line "unsubscribe target-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
