diff mbox

nvme-scsi: Use correct byte ordering for eui64 in Dev ID VPD

Message ID 20170425201809.1683-1-jonathan.derrick@intel.com (mailing list archive)
State Not Applicable, archived
Headers show

Commit Message

Jon Derrick April 25, 2017, 8:18 p.m. UTC
NVME specifies an EUI64/NGUID in little-endian format, while SCSI
specifies that the Device Identification VPD use big-endian for EUI
formats. The current code copies this bytestream directly from the
Identification Namespace page, meaning we just need to reverse the
bytestream when passing it on to the VPD.

This results in the correct values being parsed by sg-tools, eg:
$ sg_inq -e -p 0x83 /dev/nvme1n1 -vvv
open /dev/nvme1n1 with flags=0x800
VPD INQUIRY: Device Identification page
    inquiry cdb: 12 01 83 00 fc 00
      duration=0 ms
  Designation descriptor number 1, descriptor length: 20
    designator_type: EUI-64 based,  code_set: Binary
    associated with the addressed logical unit
      EUI-64 based 16 byte identifier
      Identifier extension: 0x100000001
      IEEE Company_id: 0x5cd2e4
      Vendor Specific Extension Identifier: 0x0
      [0x00000001000000015cd2e40000000000]

Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
---
 drivers/nvme/host/scsi.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

Comments

Jon Derrick April 25, 2017, 8:55 p.m. UTC | #1
On 04/25/2017 02:18 PM, Jon Derrick wrote:
> NVME specifies an EUI64/NGUID in little-endian format, while SCSI
> specifies that the Device Identification VPD use big-endian for EUI
> formats. The current code copies this bytestream directly from the
> Identification Namespace page, meaning we just need to reverse the
> bytestream when passing it on to the VPD.
> 

This seems to hold true for NGUID devices, but Keith just pointed out to
me that it may not hold true for EUI64 devices. It seems like that case
needs byte swiveling within each field. So I'll NAK for now until I can
figure out if that's the case.
Christoph Hellwig April 26, 2017, 7:22 a.m. UTC | #2
On Tue, Apr 25, 2017 at 02:18:09PM -0600, Jon Derrick wrote:
> NVME specifies an EUI64/NGUID in little-endian format, while SCSI
> specifies that the Device Identification VPD use big-endian for EUI
> formats. The current code copies this bytestream directly from the
> Identification Namespace page, meaning we just need to reverse the
> bytestream when passing it on to the VPD.

This will break existing setups that rely on VPD 0x83 for device
identification (which I think includes older SuSE distros).

And once you change the setup anyway please stop using this buggy
SCSI emulation.
diff mbox

Patch

diff --git a/drivers/nvme/host/scsi.c b/drivers/nvme/host/scsi.c
index f49ae27..8f94e77 100644
--- a/drivers/nvme/host/scsi.c
+++ b/drivers/nvme/host/scsi.c
@@ -598,6 +598,7 @@  static int nvme_fill_device_id_eui64(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 	int nvme_sc, res;
 	size_t len;
 	void *eui;
+	int i;
 
 	nvme_sc = nvme_identify_ns(ns->ctrl, ns->ns_id, &id_ns);
 	res = nvme_trans_status_code(hdr, nvme_sc);
@@ -628,7 +629,8 @@  static int nvme_fill_device_id_eui64(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 	inq_response[5] = 0x02;	/* PIV=0b | Asso=00b | Designator Type=2h */
 	inq_response[6] = 0x00;	/* Rsvd */
 	inq_response[7] = len;	/* Designator Length */
-	memcpy(&inq_response[8], eui, len);
+	for (i = 0; i < len; i++)
+		inq_response[8 + i] = ((char *)eui)[len - (i + 1)];
 
 	res = nvme_trans_copy_to_user(hdr, inq_response, alloc_len);
 out_free_id: