Bug 1065916

Sounds like the system tried to flush I/O to the non-existent disk.
I'm assuming you did not perform the dd operation with conv=direct ?

Although, if this is the case I wouldn't know why the os is ok with unplugging the disk without cleaning the page cache of any pending I/O.

Hi Ayal,
IMPORTANT part that i realized when i tried to perform the dd operation with conv=noerror,sync is:

re-description of problem:
setup:
- first vm with 1 disk and snapshot
- second vm with 3 disks:
  
  * (diskA) - bootable disk (diskA)
  * (diskB) - a disk that is the snapshot disk of the first vm attached to it 
  * (diskC) - a disk that will be the dd destination disk .

when perform 'dd' from (diskB) to (diskC) and after the operation is done, deactivates the destination disk (diskC), the vm enters into 'Paused' status due to storage I/O problem.

- And about your question if it is still relevant:
  it doesn't matter if i use conv=noerror,sync or not

(In reply to ratamir from comment #2)
> Hi Ayal,
> IMPORTANT part that i realized when i tried to perform the dd operation with
> conv=noerror,sync is:
> 
> re-description of problem:
> setup:
> - first vm with 1 disk and snapshot
> - second vm with 3 disks:
>   
>   * (diskA) - bootable disk (diskA)
>   * (diskB) - a disk that is the snapshot disk of the first vm attached to
> it 
>   * (diskC) - a disk that will be the dd destination disk .
> 
> when perform 'dd' from (diskB) to (diskC) and after the operation is done,
> deactivates the destination disk (diskC), the vm enters into 'Paused' status
> due to storage I/O problem.
> 
> - And about your question if it is still relevant:
>   it doesn't matter if i use conv=noerror,sync or not

This description bares almost no relation to the initial description.
First, I have no idea how you reached 'conv=noerror,sync' or what its relevance here is.
You need to either run dd if=... of=... bs=1M conv=direct
or run 'sync' when the dd is finished (before you disconnect the disk).
The 'sync' command is very different than what conv=sync does so that might be what through you off.
After running 'sync' detach the disk and check to see if VM pauses.

Second, you never mentioned anything about 2 VMs, only 1.  It is not clear to me why you have 2 here.  Does this reproduce if you simply have 1 VM, with 2 *raw* disks, no snapshots?

Hi Ayal,
For the first part of the comment, i used this:
# dd if=/dev/vdc of=/dev/vdb bs=1M conv=direct
and got this as output:
dd: invalid conversion: 'direct'

for the second part:
As i wrote in comment 3, i realized that the scenario is different from the original one. so the reason i have 2 vms is that i need to reproduce a 
'Backup api' scenario (i.e. snapshot disk of the first vm is attached to a second vm) in which, the attached disk is the source device for dd.
And the second vm has more disk for dd destination device.

This isn't reproduce with only 1 vm, The source device should be an attached snapshot disk.

(In reply to ratamir from comment #4)
> Hi Ayal,
> For the first part of the comment, i used this:
> # dd if=/dev/vdc of=/dev/vdb bs=1M conv=direct
> and got this as output:
> dd: invalid conversion: 'direct'

right, it's oflag=direct

> 
> for the second part:
> As i wrote in comment 3, i realized that the scenario is different from the
> original one. so the reason i have 2 vms is that i need to reproduce a 
> 'Backup api' scenario (i.e. snapshot disk of the first vm is attached to a
> second vm) in which, the attached disk is the source device for dd.
> And the second vm has more disk for dd destination device.
> 
> This isn't reproduce with only 1 vm, The source device should be an attached
> snapshot disk.

ok, so:
1. please run sync after the dd and retest.
2. are you detaching the *source* disk (the one you're reading from) or the destination disk?

Ok, i will retest it

And i'm not detaching --> deactivating the destination disk.

The 'oflag=direct' and syncing after the dd command solved the problem

Closing NOTABUG based on comment 7.