Skip to contentSkip to conversation

revert async probing of VMBus scsi device

PATCH In review Stephen Hemminger 4 participants 10 messages Jun 5, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 revert async probing of VMBus scsi device

Doing asynchronous probing can lead to reordered device names which is leads to failed mounts.

Fixes af0a5646cb8d ("use the new async probing feature for the hyperv drivers")
---
 drivers/scsi/storvsc_drv.c | 3 ---
 1 file changed, 3 deletions(-)
drivers/scsi/storvsc_drv.c +0 -3
index 8472de1007ff..56dcaa43b652 100644
@@ -1942,9 +1942,6 @@ static struct hv_driver storvsc_drv = {
 	.id_table = id_table,
 	.probe = storvsc_probe,
 	.remove = storvsc_remove,
-	.driver = {
-		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
-	},
 };
 
 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
Signature · 1 line
2.20.1
maintainer Reviewed-by on v1 1/1
#1 lore raw
Quoted 11 lines

-----Original Message-----
From: [email protected] <linux-hyperv- [email protected]> On Behalf Of Stephen Hemminger
Sent: Wednesday, June 5, 2019 2:52 PM
To: [email protected]
Cc: [email protected]; Stephen Hemminger <[email protected]>
Subject: [PATCH] revert async probing of VMBus scsi device

Quoted 7 lines

Doing asynchronous probing can lead to reordered device names which is leads to failed mounts.

Fixes af0a5646cb8d ("use the new async probing feature for the hyperv

drivers")

---

#2 lore raw
In reply to Stephen Hemminger
Quoted 2 lines · Stephen Hemminger

Doing asynchronous probing can lead to reordered device names which is leads to failed mounts.

Which is true for every device, and why we use UUIDs or label for mounts that are supposed to be stable.

#3 lore raw
Quoted 6 lines · On Wed, 5 Jun 2019 11:56:37 -0700 Christoph Hellwig <[email protected]> wrote:
Quoted 2 lines · Stephen Hemminger

Doing asynchronous probing can lead to reordered device names which is leads to failed mounts.

Which is true for every device, and why we use UUIDs or label for mounts that are supposed to be stable.

Not everyone is smart enough to do that.

#4 lore raw
Quoted 4 lines · Stephen Hemminger
Quoted 2 lines

Which is true for every device, and why we use UUIDs or label for mounts that are supposed to be stable.

Not everyone is smart enough to do that.

Sure. But they should not get a way out for just one specific driver.

#5 lore raw
Quoted 7 lines · On Wed, 5 Jun 2019 12:07:23 -0700 Christoph Hellwig <[email protected]> wrote:
Quoted 4 lines · Stephen Hemminger
Quoted 2 lines

Which is true for every device, and why we use UUIDs or label for mounts that are supposed to be stable.

Not everyone is smart enough to do that.

Sure. But they should not get a way out for just one specific driver.

There are people running new kernels on 6 year old distributions. Was every distribution smart enough then? If you think so, then this not necessary.

#6 lore raw
Quoted 5 lines · Stephen Hemminger
Quoted 1 line

Sure. But they should not get a way out for just one specific driver.

There are people running new kernels on 6 year old distributions. Was every distribution smart enough then? If you think so, then this not necessary.

I think you are missing my point. If we want a way to disable this, we:

a) want it opt-in
b) it needs to for the whole SCSI layer and not just one driver

#7 lore raw
Quoted 7 lines · On Wed, 5 Jun 2019 12:26:47 -0700 Christoph Hellwig <[email protected]> wrote:
Quoted 5 lines · Stephen Hemminger
Quoted 1 line

Sure. But they should not get a way out for just one specific driver.

There are people running new kernels on 6 year old distributions. Was every distribution smart enough then? If you think so, then this not necessary.

Quoted 5 lines

I think you are missing my point. If we want a way to disable this, we:

a) want it opt-in
b) it needs to for the whole SCSI layer and not just one driver

There is some possible issues with how initial images are deployed with temporary disks. The temp disks come pre-formatted with a blank NTFS and cloudinit reformats them to ext4 the first time.

Not sure if ordering matters, or if cloudinit is smart enough to do the right thing. I am trying to get an answer.

It might just be that the first boot should turn off async probing for the whole scsi layer via kernel cmdline. Or it might not be an issue if cloudinit was written by developers smart enough to handle moving disks.

#8 lore raw
In reply to Christoph Hellwig
Quoted 7 lines · On Wed, 5 Jun 2019 12:26:47 -0700 Christoph Hellwig <[email protected]> wrote:
Quoted 5 lines · Stephen Hemminger
Quoted 1 line

Sure. But they should not get a way out for just one specific driver.

There are people running new kernels on 6 year old distributions. Was every distribution smart enough then? If you think so, then this not necessary.

Quoted 5 lines

I think you are missing my point. If we want a way to disable this, we:

a) want it opt-in
b) it needs to for the whole SCSI layer and not just one driver

Based on feedback from the Program Manager who handles the distros the patch for storvsc is not needed.

SCSI device naming was never deterministic anyway.
Cloud-init relies on the same mechanism as the Azure agent to detect the OS and ephemeral disk.

#9 lore raw
In reply to Christoph Hellwig

From: Christoph Hellwig <[email protected]>
Date: Wed, 5 Jun 2019 12:07:23 -0700

Quoted 7 lines · Stephen Hemminger
Quoted 4 lines
Quoted 2 lines

Which is true for every device, and why we use UUIDs or label for mounts that are supposed to be stable.

Not everyone is smart enough to do that.

Sure. But they should not get a way out for just one specific driver.

Agreed.