[linux-lvm] Sector size is assumed 512

[linux-lvm] Sector size is assumed 512

[Tomasz Lasko]

[-- Attachment #1: Type: text/plain, Size: 939 bytes --]

Hi,

I'm not a part of the list or the project, just a random guy dropping by 
to say I found one suspicious thing:

after looking for what 's' size stands for, I found that your 
lvmcmdline.c source code 
<https://github.com/Jajcus/lvm2/blob/master/tools/lvmcmdline.c#L320> 
probably assumes that sector size is 512, but there are various sector 
sizes out in world (both for the hardware sector size and logical disk 
interfaces like SCSI) especially more and more popular 4096 byte sector 
size.

I wonder if apart of lvmcmdline.c above, also other parts of your 
software assume that sector size will always be 512. If yes, then I 
suggest rethinking if 4k sectors might break some operations in LVM.

By the way, I understand that when specifying command line parameter 
sector size, then it is the same for small 's' as capital 'S', right?  
And the same goes for bytes ('b' is the same as 'B'), right?

Best regards,
Tomasz Lasko



[-- Attachment #2: Type: text/html, Size: 1257 bytes --]

Re: [linux-lvm] Sector size is assumed 512

[Alasdair G Kergon]

It's been a linux kernel coding standard to treat the sector size as
always 512 bytes which we have adopted - regardless of the actual
hardware underneath.

S and s, and B and b mean the same.

Both userspace and kernel code automatically respects alignment
constraints imposed by the devices underneath, such as larger sector sizes.

Alasdair

Re: [linux-lvm] Sector size is assumed 512

[Zdenek Kabelac]

Dne 11.5.2017 v 12:39 Tomasz Lasko napsal(a):
> Hi,
> 
> I'm not a part of the list or the project, just a random guy dropping by to 
> say I found one suspicious thing:
> 
> after looking for what 's' size stands for, I found that your lvmcmdline.c 
> source code 
> <https://github.com/Jajcus/lvm2/blob/master/tools/lvmcmdline.c#L320> probably 
> assumes that sector size is 512, but there are various sector sizes out in 
> world (both for the hardware sector size and logical disk interfaces like 
> SCSI) especially more and more popular 4096 byte sector size.
> 
> I wonder if apart of lvmcmdline.c above, also other parts of your software 
> assume that sector size will always be 512. If yes, then I suggest rethinking 
> if 4k sectors might break some operations in LVM.
> 
> By the way, I understand that when specifying command line parameter sector 
> size, then it is the same for small 's' as capital 'S', right?  And the same 
> goes for bytes ('b' is the same as 'B'), right?
> 

Hi

Sector size is 'fixed'  for  dm devices for 512b.

If you work with 4K sector disks you need to use/specify size multiplied by 8.

It would get really messy if you would have multiple different PVs in a VG 
with  different sector size and you would want to print i.e. size of LV in 
sectors.

So nope - we can't change this fixed 'base' value.

Normally users do work size in  MiB or GiB eventually TiB.

lvm2 never advises users to go to sector level precision for numerous reason.

If you care about single sector - lvm2 is likely not a tool to be considered 
for usage.


Also note - default alignment is 4MiB - so unless you would override this 
extent-size below 4K, you can't actually hit issues with wrong alignment
(and likely >99% users never change this setting)

And BTW it's explicitly mentioned that 'sector is 512b' it has not attachment 
to physical device sector size...

Regards


Zdenek