Skip to contentSkip to conversation

nvme: check for dynamic controller model before verifying cntlid

PATCH In review David Milburn 2 participants 4 messages Oct 16, 2019 – Oct 17, 2019 NVMe

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 nvme: check for dynamic controller model before verifying cntlid

If controller supports dynamic controller model the cntlid returned from the identify command maybe FFFFh causing check in nvme_init_identify() to fail which results in "nvme discover" failure.

nvme discover -t rdma -a <target_IP>
Failed to write to /dev/nvme-fabrics: Invalid argument

---
 drivers/nvme/host/core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/nvme/host/core.c +1 -1
index fd7dea36c3b6..235aa1811446 100644
@@ -2812,7 +2812,7 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)
 		 * In fabrics we need to verify the cntlid matches the
 		 * admin connect
 		 */
-		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
+		if (id->ctrattr && (ctrl->cntlid != le16_to_cpu(id->cntlid))) {
 			ret = -EINVAL;
 			goto out_free;
 		}
Signature · 7 lines
2.18.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
#1 lore raw

drivers/nvme/host/core.c int nvme_init_identify(struct nvme_ctrl *ctrl) · line 2815

Quoted 2 lines · David Milburn
drivers/nvme/host/core.c +1 -1
-		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
+		if (id->ctrattr && (ctrl->cntlid != le16_to_cpu(id->cntlid))) {

Where do get the all-Fs cntrlid? Identify Controller should return the actual controller id, and the fabrics setup code also sets it up to the return value of the connect command, which shouldn't be all-Fs either.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#2 lore raw

Hi Christoph,

Quoted 8 lines · Christoph Hellwig

drivers/nvme/host/core.c int nvme_init_identify(struct nvme_ctrl *ctrl) · line 2815

Quoted 2 lines · David Milburn
drivers/nvme/host/core.c +1 -1
-		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
+		if (id->ctrattr && (ctrl->cntlid != le16_to_cpu(id->cntlid))) {

Where do get the all-Fs cntrlid? Identify Controller should return the actual controller id, and the fabrics setup code also sets it up to the return value of the connect command, which shouldn't be all-Fs either.

I see it in nvme_init_identify() after doing the nvme_identify_ctrl(), id->cntlid is FFFF causing the check to fail. There is a discussion in the NVMe over Fabrics spec, not specifically about identify, but with discover log page concerning dynamic vs static model and the value of cntlid, the target is a storage array.

Thanks,
David

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

#3 lore raw
Quoted 12 lines · David Milburn

Hi Christoph,

Quoted 8 lines · Christoph Hellwig

drivers/nvme/host/core.c int nvme_init_identify(struct nvme_ctrl *ctrl) · line 2815

Quoted 2 lines · David Milburn
drivers/nvme/host/core.c +1 -1
-		if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
+		if (id->ctrattr && (ctrl->cntlid != le16_to_cpu(id->cntlid))) {

Where do get the all-Fs cntrlid? Identify Controller should return the actual controller id, and the fabrics setup code also sets it up to the return value of the connect command, which shouldn't be all-Fs either.

Quoted 5 lines

I see it in nvme_init_identify() after doing the nvme_identify_ctrl(), id->cntlid is FFFF causing the check to fail. There is a discussion in the NVMe over Fabrics spec, not specifically about identify, but with discover log page concerning dynamic vs static model and the value of cntlid, the target is a storage array.

Please trace down where it comes from, because I suspect your array is buggy. I guess it returns 0xffff from the connect command response, which is wrong as that should contain the actual controller id, and not 0xffff which is passed in the connect command to request the dynamic controller model.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme