On 26.5.2016 16:12, Mikulas Patocka wrote:


On Thu, 26 May 2016, Navin P.S wrote:

Hi, Can you please help me with the below query that i posted to dm-devel ?

On Thu, May 26, 2016 at 5:07 AM, Navin P.S <[email protected]> wrote:
      Hi,  If i have a snapshot origin created using dmsetup create table with 
a mapping  real device  with (major,minor) =(8,18) and lets name it 
/dev/mapper/dmso
      . Now the set of reads are writes done from /dev/mapper/dmso have some 
values (R,W).

Had i been creating a persistent snapshot using (P or PO) mode with origin 
device as /dev/null and COW device is my real device (maj,min)=8,18 and it is 
named as
/dev/mapper/dmsp. Now if i do the exact same reads and writes that i did to 
dmso (snapshot-origin) , will get i the same set of output values (R,W).

In my experiment i did. So i'm asking is there a case where it differs ?

Also i looked at the code in drivers/md/dm-snap.c and i find that both the 
drivers snapshot and snapshot-origin use the same __orgin_write . So my
partial/incomplete understanding is that it works from the read/write view 
point from /dev/mapper/dmsp.

I'll not be using the snapshot-merge feature , so merging to /dev/null is 
irrelevant here.

Regards,
-- Navin

Hi

The snapshot target doesn't accept /dev/null because it requires a block
device and /dev/null is chatected device.

You can use the "zero" device mapper target to create a block device that
returns zeroes on reads and that discards writes. You can use this zero
device as an origin to the dm-snapshot target.

With LVM, you can use this functionality with lvcreate --virtualsize
switch.



Just a minor clarification - if you have any 'recent' version of lvm2
this functionality is handled via thin-provisioning by default.

So if you really want to use 'old' snapshot target - just use:

lvcreate --type snapshot -Lrealsize  -Vvirtualsize  vg


And if you want to see exact ioctl sequencing - then use lvcreate -vvvv.

Regards

Zdenek


--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel

Reply via email to