qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] How to discard one range which overlap with backing fil


From: Eric Blake
Subject: Re: [Qemu-devel] How to discard one range which overlap with backing file and its children img?
Date: Tue, 7 Aug 2018 09:02:24 -0500
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.8.0

On 08/07/2018 07:34 AM, Kevin Wolf wrote:
Am 07.08.2018 um 09:06 hat lampahome geschrieben:
I have image A & B, and A is backing file of B.

After I mount A to /dev/nbd0 and I write from position 0~999 in nbd0.

Then create B and set A as backing file of B.

I mount B on /dev/nbd1 and I can saw the data from pos:0~999 because A is
B's backing file. That's reasonable.


But I want to discard range 0~500 in B. I expect there's no data in 0~500
after discard and re-mount B next time.

But the data is still in A.

How can I discard range 0~500?

Note that discard simply means that you don't care about the content any
more. This doesn't guarantee that the old data can't be read any more.
If you want to make the data invisible, you need a zero write operation
instead. For a Linux guest, have a look at the "fallocate -z" command
line tool.

What's more, once you make A the backing file to B, then all actions you perform through qemu operate only on B (A is treated as read-only). So, writing zeroes (or discarding) changes the contents of B, but does NOT change the content of A.

Furthermore, both discard and write zeroes tend to have alignment constraints. If you are actually working on bytes 0-999 (which is unaligned to sector boundaries of 512, let alone qcow2 cluster boundaries that default to 64k but can be even larger), the discard is likely to be a no-op (since discard tends to be a no-op if it does not meet larger alignment boundaries), while the write zeroes will work but will result in a slower read-modify-write if it is targetting a block device. If you are instead working on sectors 0-999 (bytes 0-511999), you have more of a chance of seeing discard in action, as well as write zeroes being more efficient.

--
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org



reply via email to

[Prev in Thread] Current Thread [Next in Thread]