Copying volumes to storage via Linux server using XCOPY

It happens that you need to get a complete copy of a volume within one data storage system (SDS), not a snapshot, a clone, but a full-fledged volume. But storage does not always allow this to be done within itself with its own means. It seems that the only option is to copy through the server, but at the same time, the entire amount of data will be driven through the server itself, the network to the storage system and the storage ports, loading all these components. But there are SCSI commands that can allow you to do everything within the storage system itself, and if your system supports VAAI from VMware, then it is almost 100% that the XCOPY (EXTENDED COPY) command is supported, which tells the array what and where to copy, without involving in this process server and network.

It seems like everything should be simple, but I didn’t find ready-made scripts right off the bat, I had to reinvent the wheel. Linux was chosen for the server OS, and the ddpt command (http://sg.danny.cz/sg/ddpt.html) was chosen as the copying tool. Using this combination, you can copy any volumes from any OS, since copying occurs block by block on the storage side. Since it is necessary to copy block by block, and the number of blocks must be counted, the blockdev command was used to count the number of such iterations. The maximum block size was obtained empirically, with a large block, ddpt did not work in fact. The result is the following rather simple script:

#!/bin/bash
# first parameter = input device
# second parameter = output device
# device size must be the same
# changing bs variable can reduce speed, max speed should be at bs=32768. 32768 is max setting, lower settings should be calculated dividing by 2

set -o nounset
bs=32768
s=`blockdev --getsz $1`
i=0
while [ $i -le $s ]
do
ddpt of=$2 bs=512 oflag=xcopy,direct if=$1 iflag=xcopy,direct count=$bs verbose=-1 skip=$i seek=$i
i=$(( $i+$bs ))
done

Let's do a little check! Well, as a small, 1TB file was created and checked by md5sum slowly 🙂

root@sales-demo-05:/home/vasilyk# blockdev --getsz /dev/mapper/mpathfs
2516582400
root@sales-demo-05:/home/vasilyk# blockdev --getsz /dev/mapper/mpathfr
2516582400
root@sales-demo-05:/home/vasilyk# mount /dev/mapper/mpathfs /xcopy_source/
mount: /xcopy_source: wrong fs type, bad option, bad superblock on /dev/mapper/mpathfs, missing codepage or helper program, or other error.
root@sales-demo-05:/home/vasilyk# mkfs /dev/mapper/mpathfs
mke2fs 1.44.1 (24-Mar-2018)
Discarding device blocks: done
Creating filesystem with 314572800 4k blocks and 78643200 inodes
Filesystem UUID: bed3ea00-c181-4b4e-b52e-d9bb498be756
Superblock backups stored on blocks:
        32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
        4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968,
        102400000, 214990848

Allocating group tables: done
Writing inode tables: done
Writing superblocks and filesystem accounting information: done

root@sales-demo-05:/home/vasilyk# mount /dev/mapper/mpathfs /xcopy_source/
root@sales-demo-05:/home/vasilyk# ls -l /xcopy_source/
total 16
drwx------ 2 root root 16384 Aug 19 15:35 lost+found
root@sales-demo-05:/home/vasilyk# head -c 1T </dev/urandom > /xcopy_source/1TB_file
root@sales-demo-05:/home/vasilyk# ls -l /xcopy_source/
total 1074791444
-rw-r--r-- 1 root root 1099511627776 Aug 19 17:25 1TB_file
drwx------ 2 root root         16384 Aug 19 15:35 lost+found
root@sales-demo-05:/home/vasilyk# umount /xcopy_source
root@sales-demo-05:/home/vasilyk# mount /dev/mapper/mpathfr /xcopy_dest/
mount: /xcopy_dest: wrong fs type, bad option, bad superblock on /dev/mapper/mpathfr, missing codepage or helper program, or other error.
root@sales-demo-05:/home/vasilyk# cat xcopy.sh
#!/bin/bash
# first parameter = input device
# second parameter = output device
# device size must be the same
# changing bs variable can reduce speed, max speed should be at bs=32768. 32768 is max setting, lower settings should be calculated dividing by 2

bs=32768
s=`blockdev --getsz $1`
i=0
while [ $i -le $s ]
do
ddpt of=$2 bs=512 oflag=xcopy,direct if=$1 iflag=xcopy,direct count=$bs verbose=-1 skip=$i seek=$i
i=$(( $i+$bs ))
done
root@sales-demo-05:/home/vasilyk# time ./xcopy.sh /dev/mapper/mpathfs /dev/mapper/mpathfr
real    11m30.878s
user    2m3.000s
sys     1m11.657s

What was happening on the storage system at that moment:

Let's continue with Linux.

root@sales-demo-05:/home/vasilyk# mount /dev/mapper/mpathfr /xcopy_dest/
root@sales-demo-05:/home/vasilyk# ls -l /xcopy_dest/
total 1074791444
-rw-r--r-- 1 root root 1099511627776 Aug 19 17:25 1TB_file
drwx------ 2 root root         16384 Aug 19 15:35 lost+found
root@sales-demo-05:/home/vasilyk# mount /dev/mapper/mpathfs /xcopy_source/
root@sales-demo-05:/home/vasilyk# md5sum /xcopy_source/1TB_file
53dc6dfdfc89f099c0d5177c652b5764  /xcopy_source/1TB_file
root@sales-demo-05:/home/vasilyk# md5sum /xcopy_dest/1TB_file
53dc6dfdfc89f099c0d5177c652b5764  /xcopy_dest/1TB_file
root@sales-demo-05:/home/vasilyk#

Everything worked out, but test and use at your own peril and risk! As a source volume, it is better to take snapshots, for starters.

Source: habr.com