EBS Volumes get stuck detaching from AWS instances

[Matthew Ruffell]

On AWS, it is possible to get a EBS volume stuck in the detaching state, where it is not present in lsblk or lspci, but the cloud console says the volume is detaching, and pci / nvme errors are output to dmesg every 180 seconds.

To reproduce reliably:

1) Start a AWS instance. Can be any type, nitro or regular. I have successfully reproduced on m5.large and t3.small. Add an extra volume during creation, of any size.

2) Connect to the instance. lsblk and lspci show the nvme device there.
Detach the volume from the web console. It will detach successfully.
You can attach and then detach the volume again, and it will also work
successfully.

3) With the volume detached, reboot the instance. I do "sudo reboot".

4) When the instance comes back up and you have logged in, attach the volume.

dmesg will have (normal):

kernel: [ 67] pci 0000:00:1f.0: [1d0f:8061] type 00 class 0x010802
kernel: [ 67] pci 0000:00:1f.0: reg 0x10: [mem 0x00000000-0x00003fff]
kernel: [ 67] pci 0000:00:1f.0: BAR 0: assigned [mem 0x80000000-0x80003fff]
kernel: [ 67] nvme nvme1: pci function 0000:00:1f.0
kernel: [ 67] nvme 0000:00:1f.0: enabling device (0000 -> 0002)
kernel: [ 67] ACPI: PCI Interrupt Link [LNKC] enabled at IRQ 10

5) Detach the volume from the web console. If you keep refreshing the volume
view, you will see the volume in the detaching state and the volume is still
in use.

The device will be missing from lsblk and lspci.

dmesg will print these messages every 180 seconds:

4.4 -> 4.15

kernel: [ 603] pci 0000:00:1f.0: [1d0f:8061] type 00 class 0x010802
kernel: [ 603] pci 0000:00:1f.0: reg 0x10: [mem 0x80000000-0x80003fff]
kernel: [ 603] pci 0000:00:1f.0: BAR 0: assigned [mem 0x80000000-0x80003fff]
kernel: [ 603] nvme nvme1: pci function 0000:00:1f.0
kernel: [ 603] nvme nvme1: failed to mark controller live
kernel: [ 603] nvme nvme1: Removing after probe failure status: 0

Latest mainline kernel:

kernel: [ 243] pci 0000:00:1f.0: [1d0f:8061] type 00 class 0x010802
kernel: [ 243] pci 0000:00:1f.0: reg 0x10: [mem 0xc0000000-0xc0003fff]
kernel: [ 243] pci 0000:00:1f.0: BAR 0: assigned [mem 0xc0000000-0xc0003fff]
kernel: [ 243] nvme nvme1: pci function 0000:00:1f.0
kernel: [ 244] nvme nvme1: failed to mark controller CONNECTING
kernel: [ 244] nvme nvme1: Removing after probe failure status: 0

The volume is now stuck detaching and the above will continue until you
forcefully detach the volume.

This seems to effect all distributions and all kernels.
I have tested xenial, bionic, bionic + hwe, bionic + eoan,
bionic + mainline 5.2.2, rhel8 and Amazon Linux 2. All are affected with the
same symptoms.

I tried to trace NVMe on 5.2, but with not much success on finding any problem:
https://paste.ubuntu.com/p/c6rmDpvHJk/

[Ubuntu Kernel Bot (ubuntu-kernel-bot)]

This bug is missing log files that will aid in diagnosing the problem. While running an Ubuntu kernel (not a mainline or third-party kernel) please enter the following command in a terminal window:

apport-collect 1837833

and then change the status of the bug to 'Confirmed'.

If, due to the nature of the issue you have encountered, you are unable to run this command, please add a comment stating that fact and change the bug status to 'Confirmed'.

This change has been made by an automated script, maintained by the Ubuntu Kernel Team.

[Johannes Würbach (johannes-wuerbach)]

@mruffell we saw the same issue on our AWS machines, but a kernel version including https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=linux-4.19.y&id=f0c83dd15ee1e89f73523cb82da9205d204cf440 (e.g. 4.19.61) fixed it for us.

[Matthew Ruffell (mruffell)]

Hello @johannes-wuerbach

Thanks for the suggestion. I had a look at the patch you suggested:

commit 2181e455612a8db2761eabbf126640552a451e96
Author: Anton Eidelman <email address hidden>
Date: Thu Jun 20 08:48:10 2019 +0200
subject: nvme: fix possible io failures when removing multipathed ns

The patch was backported to 5.2.3, 5.1.20 and 4.19.61, so I went and tested it.

I did a quick test by booting mainline 5.2.2, reproduced the issue, and then installed 5.2.3. I was able to reproduce the issue again, suggesting that this commit does not fix the issue.

I then tried the latest 5.2.8, and also reproduced the issue.

Just to be sure, I backported the commit to 4.15.0-58-generic, and tested again:

You can find the test kernel here:

https://launchpad.net/~mruffell/+archive/ubuntu/sf234512-test

I was also able to reproduce the issue, and I am confident that this commit does nothing for this particular problem.

Thanks for the suggestion! But in this particular case that commit is not the answer.

I have opened a ticket with AWS, and they have confirmed that they know about the problem and have developed a fix for their cloud platform. It is currently being deployed, and deployment will take a few weeks to complete.

You might have been lucky and received a patched cloud machine at the same time 4.19.61 came out, or it fixed another unrelated issue that had the same symptoms.

[Matthew Ruffell (mruffell)]

I have received confirmation from AWS that the fix for this issue has been
deployed to all commercial regions.

I have tested multiple VMs over the last few days and detaching EBS volumes are
working without problems.

This issue is now resolved.

[test0830 (test0830)]

Hello @mruffell , How to find this patch for this problem?
thanks

[Matthew Ruffell (mruffell)]

Hi test0830,

This particular issue was fixed by AWS on their backend, it wasn't an issue with Ubuntu at all.

If you are having similar issues, you should open a new bug or a support ticket on AWS.

Thanks,
Matthew