Am 20.01.2025 um 23:21 hat Fahrzin Hemmati geschrieben:
Happy to wait until your patchset is in.
Yes, this is a no-op, but it reads the entire disk image to perform that
no-op, so this is merely a time-saving improvement, not a behavior change.
Can you give more context on what exactly you're doing that it reads the
entire disk image just to resize it? This sounds completely unexpected
and if it does, there may be a different problem to be solved.
In my test, for a raw image, I see a single ftruncate() call, which is
unnecessary, but shouldn't cause any measurable time difference for the
qemu-img run. qcow2 has a little more code in QEMU to figure out that
there is nothing to do, but it doesn't involve any syscall and certainly
not reading the whole image.
Kevin
On Mon Jan 20, 2025, 09:49 PM GMT, Michael Tokarev <mjt@tls.msk.ru> wrote:
20.01.2025 23:17, Fahrzin Hemmati wrote:
My apologies, I saw the Signed-off-by requirement at first, but as I
followed the docs I got to "git publish" and didn't realize "git publish"
was
actually able to send emails on my system (I don't remember setting up any
SMTP settings). By that time, I forgot and thought this patch was short
enough to not warrant much of a commit message.
The main practical advantage is for users that call "qemu-img resize" via
scripts or other systems (like Packer in my case) that don't check the
image
size ahead of time. I can upstream this change into them (by using
"qemu-img info —output=json ...") but I figured it would be useful to more
users here.
This could trivially be added to block/io.c:bdrv_co_truncate(), as well as
blockdev.c:qmp_block_resize() with a little more work. I'm not familiar
with those workflows, but if needed I can do that as well.
Here's the new patch:
From 17f5c5f03d930c4816b92b97e0e54db0725d7b94 Mon Sep 17 00:00:00 2001
From: Fahrzin Hemmati <fahhem@fahhem.com <mailto:fahhem@fahhem.com>>
Date: Mon, 20 Jan 2025 01:56:24 -0800
Subject: [PATCH] Skip resizing image to the same size
Higher-level software, such as Packer, blindly call "qemu-img resize"
even when the size is the same. This change can be pushed to those
users, but it's very cheap to check and can save many other users more
time here.
Signed-off-by: Fahrzin Hemmati <fahhem@fahhem.com <mailto:fahhem@fahhem.com
>>
---
qemu-img.c | 6 ++++++
tests/qemu-iotests/061 | 8 ++++++++
tests/qemu-iotests/061.out | 9 +++++++++
3 files changed, 23 insertions(+)
diff --git a/qemu-img.c b/qemu-img.c
index 2f2fac90e8..3345c4e63f 100644
--- a/qemu-img.c
+++ b/qemu-img.c
@@ -4184,6 +4184,12 @@ static int img_resize(int argc, char **argv)
goto out;
}
+ if (total_size == current_size) {
+ qprintf(quiet, "Image already has the desired size.\n");
+ ret = 0;
+ goto out;
+ }
I don't think this is necessary: the actual operation is a no-op
anyway, there's no need to special-case it.
Please also refrain from changing qemu-img until my refresh
patchset is either accepted or rejected. It was a large work
and it'd be sad to lose it.
Thanks,
/mjt