Bug in BLKBSZSET/GET ioctl ?

Bug in BLKBSZSET/GET ioctl ?

[Wanlong Gao]

Hi all:

This is first reported to *libguestfs*: https://bugzilla.redhat.com/show_bug.cgi?id=624335

Then, I looked into upstream util-linux and it seems nothing wrong. I'm not convinced that it's a kernel bug.

produce:

---
# ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
4096
# ./util-linux/disk-utils/blockdev --setbsz 2048 /dev/sda6
# ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
4096
# ./util-linux/disk-utils/blockdev --setbsz 512 /dev/sda6
# ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
4096

---

I'm not sure if it is related to write-back, but still CCed to Fengguang.



Thanks,
-Wanlong Gao

Re: Bug in BLKBSZSET/GET ioctl ?

[Wanlong Gao]

On 12/29/2011 11:51 AM, Wanlong Gao wrote:

> Hi all:
> 
> This is first reported to *libguestfs*: https://bugzilla.redhat.com/show_bug.cgi?id=624335


The blockdev-get/setbsz in *libguestfs* is just a wraper of util-linux's *blockdev --get/setbsz*.

> 
> Then, I looked into upstream util-linux and it seems nothing wrong. I'm not convinced that it's a kernel bug.
> 
> produce:
> 
> ---
> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> 4096
> # ./util-linux/disk-utils/blockdev --setbsz 2048 /dev/sda6
> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> 4096
> # ./util-linux/disk-utils/blockdev --setbsz 512 /dev/sda6
> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> 4096
> 
> ---
> 
> I'm not sure if it is related to write-back, but still CCed to Fengguang.
> 
> 
> 
> Thanks,
> -Wanlong Gao

Re: Bug in BLKBSZSET/GET ioctl ?

[Richard W.M. Jones]

On Thu, Dec 29, 2011 at 11:56:27AM +0800, Wanlong Gao wrote:
> On 12/29/2011 11:51 AM, Wanlong Gao wrote:
> 
> > Hi all:
> > 
> > This is first reported to *libguestfs*: https://bugzilla.redhat.com/show_bug.cgi?id=624335
> 
> 
> The blockdev-get/setbsz in *libguestfs* is just a wraper of util-linux's *blockdev --get/setbsz*.

Agreed.

I don't think libguestfs is anything to do with it, since those
commands simply translate to running
/sbin/blockdev --{get,set}bsz /dev/sdX:

http://git.annexia.org/?p=libguestfs.git;a=blob;f=daemon/blockdev.c;h=d056abef1739de7b9e9aab118c138bae403e11e5;hb=HEAD#l106

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
virt-top is 'top' for virtual machines.  Tiny program with many
powerful monitoring features, net stats, disk stats, logging, etc.
http://et.redhat.com/~rjones/virt-top

Re: Bug in BLKBSZSET/GET ioctl ?

[Wu Fengguang]

On Thu, Dec 29, 2011 at 11:51:45AM +0800, Wanlong Gao wrote:
> Hi all:
> 
> This is first reported to *libguestfs*: https://bugzilla.redhat.com/show_bug.cgi?id=624335
> 
> Then, I looked into upstream util-linux and it seems nothing wrong. I'm not convinced that it's a kernel bug.
> 
> produce:
> 
> ---
> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> 4096
> # ./util-linux/disk-utils/blockdev --setbsz 2048 /dev/sda6
> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> 4096
> # ./util-linux/disk-utils/blockdev --setbsz 512 /dev/sda6
> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> 4096

I think each blockdev invocation is working on a *new* bdev object.
You'll get consistent results if somehow keep it referenced, for
example, run a background copy task:

# cp /dev/sda6 /dev/null &

# ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
# ./util-linux/disk-utils/blockdev --setbsz 512 /dev/sda6
# ./util-linux/disk-utils/blockdev --getbsz /dev/sda6

Thanks,
Fengguang

Re: Bug in BLKBSZSET/GET ioctl ?

[Wanlong Gao]

On 12/29/2011 12:20 PM, Wu Fengguang wrote:

> On Thu, Dec 29, 2011 at 11:51:45AM +0800, Wanlong Gao wrote:
>> Hi all:
>>
>> This is first reported to *libguestfs*: https://bugzilla.redhat.com/show_bug.cgi?id=624335
>>
>> Then, I looked into upstream util-linux and it seems nothing wrong. I'm not convinced that it's a kernel bug.
>>
>> produce:
>>
>> ---
>> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
>> 4096
>> # ./util-linux/disk-utils/blockdev --setbsz 2048 /dev/sda6
>> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
>> 4096
>> # ./util-linux/disk-utils/blockdev --setbsz 512 /dev/sda6
>> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
>> 4096
> 
> I think each blockdev invocation is working on a *new* bdev object.


But the address of *new* bdev is the same?
I did printk, and they all returned the same address.

> You'll get consistent results if somehow keep it referenced, for


But isn't it a bug? It seems that the setbsz has no effect?

Thanks,
-Wanlong

> example, run a background copy task:
> 
> # cp /dev/sda6 /dev/null &
> 
> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> # ./util-linux/disk-utils/blockdev --setbsz 512 /dev/sda6
> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> 
> Thanks,
> Fengguang
> 

Re: Bug in BLKBSZSET/GET ioctl ?

[Wu Fengguang]

On Thu, Dec 29, 2011 at 12:28:44PM +0800, Wanlong Gao wrote:
> On 12/29/2011 12:20 PM, Wu Fengguang wrote:
> 
> > On Thu, Dec 29, 2011 at 11:51:45AM +0800, Wanlong Gao wrote:
> >> Hi all:
> >>
> >> This is first reported to *libguestfs*: https://bugzilla.redhat.com/show_bug.cgi?id=624335
> >>
> >> Then, I looked into upstream util-linux and it seems nothing wrong. I'm not convinced that it's a kernel bug.
> >>
> >> produce:
> >>
> >> ---
> >> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> >> 4096
> >> # ./util-linux/disk-utils/blockdev --setbsz 2048 /dev/sda6
> >> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> >> 4096
> >> # ./util-linux/disk-utils/blockdev --setbsz 512 /dev/sda6
> >> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
> >> 4096
> > 
> > I think each blockdev invocation is working on a *new* bdev object.
> 
> 
> But the address of *new* bdev is the same?
> I did printk, and they all returned the same address.

Then the block size value should be reset in one of the bd_set_size()
calls in __blkdev_get().

> > You'll get consistent results if somehow keep it referenced, for
> 
> 
> But isn't it a bug? It seems that the setbsz has no effect?

Yeah, it does look like unexpected behavior to the end user..

Thanks,
Fengguang

Re: Bug in BLKBSZSET/GET ioctl ?

[Wanlong Gao]

On 12/29/2011 12:40 PM, Wu Fengguang wrote:

> On Thu, Dec 29, 2011 at 12:28:44PM +0800, Wanlong Gao wrote:
>> On 12/29/2011 12:20 PM, Wu Fengguang wrote:
>>
>>> On Thu, Dec 29, 2011 at 11:51:45AM +0800, Wanlong Gao wrote:
>>>> Hi all:
>>>>
>>>> This is first reported to *libguestfs*: https://bugzilla.redhat.com/show_bug.cgi?id=624335
>>>>
>>>> Then, I looked into upstream util-linux and it seems nothing wrong. I'm not convinced that it's a kernel bug.
>>>>
>>>> produce:
>>>>
>>>> ---
>>>> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
>>>> 4096
>>>> # ./util-linux/disk-utils/blockdev --setbsz 2048 /dev/sda6
>>>> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
>>>> 4096
>>>> # ./util-linux/disk-utils/blockdev --setbsz 512 /dev/sda6
>>>> # ./util-linux/disk-utils/blockdev --getbsz /dev/sda6
>>>> 4096
>>>
>>> I think each blockdev invocation is working on a *new* bdev object.
>>
>>
>> But the address of *new* bdev is the same?
>> I did printk, and they all returned the same address.
> 
> Then the block size value should be reset in one of the bd_set_size()
> calls in __blkdev_get().


Is the behaviour right that setting bd_block_size to the logical_block_size
every time when __blkdev_get()?

> 
>>> You'll get consistent results if somehow keep it referenced, for
>>
>>
>> But isn't it a bug? It seems that the setbsz has no effect?
> 
> Yeah, it does look like unexpected behavior to the end user..


And at least, the ioctl BLKBZSET seems useless.

Thanks
-Wanlong 

> 
> Thanks,
> Fengguang
>