Failure to expand shrunken resizeable image from Linux machines

[george1421]

@george1421 Well setting the image to a smaller drive than the source DID successfully replicate the OP’s issue.

jondoe@jondoe-VirtualBox ~ $ df -h
Filesystem      Size  Used Avail Use% Mounted on
udev            474M     0  474M   0% /dev
tmpfs           100M  3.6M   96M   4% /run
/dev/sda1       5.9G  5.3G  296M  95% /
tmpfs           496M  112K  496M   1% /dev/shm
tmpfs           5.0M  4.0K  5.0M   1% /run/lock
tmpfs           496M     0  496M   0% /sys/fs/cgroup
cgmfs           100K     0  100K   0% /run/cgmanager/fs
tmpfs           100M   20K  100M   1% /run/user/1000
jondoe@jondoe-VirtualBox ~ $ lsblk
NAME   MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
sr0     11:0    1 1024M  0 rom  
sda      8:0    0   25G  0 disk 
└─sda1   8:1    0    6G  0 part /
jondoe@jondoe-VirtualBox ~ $ 

Note there are missing partitions too.

[Xipher]

@george1421 The whole ‘going to a smaller disk’ thing seems to be in relation to offsets being passed that exceed the ‘physical’ dimensions of the drive being restored to, see this prior gallery for the errors related.

http://imgur.com/a/LLH0m

[Xipher]

OK, so, after a fair bit of testing and retesting, here is what I’ve found out.

First of all, the following image gallery I created of the issue explains things well:

---

http://imgur.com/a/2PAWT

---

The TL;DR version is this:

LVM breaks things something /fierce/ with Mint’s default layout. If you choose to use LVM and don’t set a custom partition layout, what you get from the 18.2 installer is whats in the screenshot and it will not expand properly.

Having the root partition anywhere but as the first and primary partition of the drive makes things break. In my case, the SWAP partition was always first in my images.

Having the root partition as the first and primary partition makes the expansion work, almost 100%, it left some space but its good enough for right now.

Important Note: In my testing tonight I ended up using FOG 1.5 RC5 since I /just/ rebuilt in tonight to do this testing. Prior testing was performed on FOG 1.4.4 which for me was producing stranger results than these.

Is this intended functionality?

[Tom Elliott]

I don’t understand. So it was your layout of how it applied the partitions to disk that caused the resizing to not happen properly? So expanding does work when laid out in a specific fashion?

I was able to replicate the results through 3 vms, one capture at 50GB, one for deploying to at 30GB, and one for deploying to at 100GB (to mimic what you described as closely as possible.)

I DID find some issues, and worked very diligently to try to fix those issues. I don’t think it has anything to do with placement of LVM, directly, rather because the extended partition was moving the LVM around and it was unable to find itself based on what was presented originally.

(See Here for more information, though not directly describing it might make sense).
https://github.com/FOGProject/fogproject/commit/635c5050904c3e29edd26151d96b7217318acf6c

I was able to fix and deploy to all devices and the devices were able to boot without an issue and all showed the “expansion” had worked as well. (Well the capture system didn’t expand, but it was using the correct layout as to how it was originally configured. – Meaning it applied back exactly what should have been since it was the exact same disk. Prior to this it was expanding a tiny bit even to the same machine.)

I’ve updated the init’s. Would you be willing to give them a shot and see if the “originals” will now start working?

wget -O /var/www/fog/service/ipxe/init.xz https://fogproject.org/inits/init.xz
wget -O /var/www/fog/service/ipxe/init_32.xz https://fogproject.org/inits/init_32.xz

I’m sorry it took so long to get anything, but hope you understand that we are very busy with our own things too.

[george1421]

@tom-elliott While I’m not using a LVM disk deploying a 50GB image to a 25GB target computer seemed to work. I still have a 1.4.4 build on my fog server. I just copied over the inits.

jondoe@jondoe-VirtualBox ~ $ lsblk
NAME   MAJ:MIN RM    SIZE RO TYPE MOUNTPOINT
sr0     11:0    1   1024M  0 rom  
sda      8:0    0     25G  0 disk 
├─sda2   8:2    0      1K  0 part 
├─sda5   8:5    0 1021.8M  0 part 
└─sda1   8:1    0     24G  0 part /
jondoe@jondoe-VirtualBox ~ $ df -h
Filesystem      Size  Used Avail Use% Mounted on
udev            474M     0  474M   0% /dev
tmpfs           100M  3.6M   96M   4% /run
/dev/sda1        24G  5.3G   18G  24% /
tmpfs           496M   92K  496M   1% /dev/shm
tmpfs           5.0M  4.0K  5.0M   1% /run/lock
tmpfs           496M     0  496M   0% /sys/fs/cgroup
cgmfs           100K     0  100K   0% /run/cgmanager/fs
tmpfs           100M   20K  100M   1% /run/user/1000
jondoe@jondoe-VirtualBox ~ $ 

[george1421]

For the next test I spun up a 50GB ubuntu image (knowing the default layout is on lvm) and captured that. We all know since the disk layout is LVM FOG will capture the lvm partition as raw even for single disk resizable. That means the LVM partition is not resizable from within FOS at this point.

Here is the ubuntu reference image layout

Here is the ubuntu image deployed to a larger hard drive (65GB target vs 50GB reference image) target computer.

Here is the results of trying to deploy a 50GB captured image to a 25GB target computer:

[Xipher]

@tom-elliott I’ll give it a shot today

Going to post up the screenshots of the output in the same style everyone else is first though to try and help explain and identify the original issue better, want to try and make sure everyone is on the same page by using the same style of output.

[Xipher]

Ok, want to break down the results in the same sort of output I’m seeing used here for the source and destination of each attempt

Destination is always a 75gb machine which is always a bigger disk. The ‘smaller disk’ restore issue I think is already identified?

This is the ‘custom’ partitioned layout I was using yesterday which caused a problem:

Source:

Destination:

This is the ‘default’ layout Mint Mate 18.2 produces from its installer when LVM is selected:

Source:

Destination:

This is the ‘working’ solution wherein the root partition is the first partition on the drive:

Source:

Destination:

Something I noticed on all the machines though was that the SWAP partition ONLY mounted on the LVM destination machine…!? All other attempts without LVM lead to the destination machine not mounting its swap, really weird…

Hope this helps understand what I’m running into and I’ll try the patched init today, though I have to run off for a bit at the moment, sorry!

[Tom Elliott]

@xipher I believe the swap mount is unrelated to where the partition actually sits, rather Linux Mint seems to set the swap by uuid. While we do try to reset the UUID, this isn’t working and I’m not overly worried about fixing the UUID for SWAP partitions anyway. (Why you ask?) Because you can edit the /etc/fstab, change the swap partition from the UUID to the actual location. In my case /dev/sda5 is the drive, so replace the UUID bit with /dev/sda5 and all works properly on reboots.

[Xipher]

@tom-elliott Sorry it took so long, I put the new init in, but the results were identical sadly :C

Also, totally agree on the SWAP partition issue, its a change I’ll make on the client to capture.

[Tom Elliott]

@xipher what do you mean identical? Of course my test boxes are using the default layout of linux mint, but all are appropriately resizing where my tests yesterday before changes failed in almost the exact same way you described.

[Tom Elliott]

Also, you did update the init first? No need to rerun installer for these tests as it will write the “bad” inits if done.

[Xipher]

@tom-elliott

Ran the two commands in your prior post as root:

wget -O /var/www/fog/service/ipxe/init.xz https://fogproject.org/inits/init.xz
wget -O /var/www/fog/service/ipxe/init_32.xz https://fogproject.org/inits/init_32.xz

Recaptured all 3 images, then casted them to the test system one after the other and recorded the results.

Didn’t re-run the installer or anything else, system wasn’t rebooted, etc, pretty much just replaced the init per the above command provided, recaptured, recast.

I could try providing a VM image of one of the systems I’m capturing from that exhibits the issue? Could provide the actual images too.

[george1421]

I was just looking at my ubuntu example and I found something not right. The image is LVM so FOG can only copy that in raw mode. So no expansion can happen (this is A reason for using standard partitions). But when I went from a 50GB original image to a 65GB target computer is grew the /boot partition to by 15GB (the difference between 50 and 65). I would have thought that the /boot partition would be a fixed size partition (as with MS Windows). I created a new LM master image based on LVM and it happened the same as with ubuntu. My original intent was to have some extra space in the physical disk so I could test some LVM expansion tools.

Linux Mint with LVM defaults.

jondoe@jondoe-VirtualBox ~ $ lsblk
NAME                MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
sr0                  11:0    1 1024M  0 rom  
sda                   8:0    0   65G  0 disk 
├─sda2                8:2    0    1K  0 part 
├─sda5                8:5    0 49.5G  0 part 
│ ├─mint--vg-swap_1 253:1    0    1G  0 lvm  [SWAP]
│ └─mint--vg-root   253:0    0 48.5G  0 lvm  /
└─sda1                8:1    0 15.5G  0 part /boot
jondoe@jondoe-VirtualBox ~ $ df -h
Filesystem                 Size  Used Avail Use% Mounted on
udev                       474M     0  474M   0% /dev
tmpfs                      100M  3.6M   96M   4% /run
/dev/mapper/mint--vg-root   48G  5.0G   41G  11% /
tmpfs                      496M  168K  496M   1% /dev/shm
tmpfs                      5.0M  4.0K  5.0M   1% /run/lock
tmpfs                      496M     0  496M   0% /sys/fs/cgroup
/dev/sda1                   15G   67M   15G   1% /boot
cgmfs                      100K     0  100K   0% /run/cgmanager/fs
tmpfs                      100M  4.0K  100M   1% /run/user/108
tmpfs                      100M   20K  100M   1% /run/user/1000
jondoe@jondoe-VirtualBox ~ $ 

[Xipher]

@george1421 Close to my findings wherein the first partition (sda1) will grow proportionally, but the rest… no go.

If you create the same layout but have the root partition where you currently have the boot partition, it will grow the root and ‘work’ as it seems it would be intended.

If you were to make a home partition in the same place, it would grow but not the boot or root… etc

[Tom Elliott]

@george1421 It only expands partitions that are “expandable” and not “fixed”. Meaning, if you have 1 ext4 partition, and it shrinks, (say boot here), it will try to expand the partitions somewhat evenly. However, this is not a perfect system, nor has it ever really been. I’ve tried for all I’m worth, but I’m only one man, sorry.

[Sebastian Roth]

Great you all have been heavily working on this. Looks like we are getting there although the initially posted disk layout (sda1=root, sda2=extended, sda5=swap) has changed. From the latest pictures posted it looks as if the expansion is still not working for this kind of layout (SWAP being on sda1).

I was just able to replicate this issue. But I didn’t have the time to find what’s causing the issue in our script:

$ cat in1.txt
label: dos
label-id: 0x77265efa
device: /dev/sda
unit: sectors

/dev/sda1 : start=        2048, size=     3940352, type=82, bootable
/dev/sda2 : start=     3942400, size=    69457920, type=5
/dev/sda5 : start=     3944448, size=     3940352, type=83
/dev/sda6 : start=     7886848, size=    65513472, type=83

$ ./procsfdisk.awk -v SECTOR_SIZE=512 -v CHUNK_SIZE=512 -v MIN_START=2048 -v action=filldisk -v target=/dev/sda -v sizePos=146800640 -v diskSize=146800640 -v fixedList=1:5 in1.txt 
ERROR: New start and size (150960130) on (/dev/sda1) is larger than the disk (146800640).
# ERROR in new partition table, quitting.
# ERROR: /dev/sda5 has an overlap.
label: dos
label-id: 0x77265efa
device: /dev/sda
unit: sectors

/dev/sda1 : start=        2048, size=     3940352, type=82, bootable
/dev/sda2 : start=     3944448, size=   142856192, type=5
/dev/sda5 : start=   142856194, size=     8103936, type=83
/dev/sda6 : start=   142856194, size=   134747648, type=83

From my point of view the output of sfdisk -d /dev/sda is much more helpful than df -h.

[Tom Elliott]

@sebastian-roth said in Failure to expand shrunken resizeable image from Linux machines:

./procsfdisk.awk -v SECTOR_SIZE=512 -v CHUNK_SIZE=512 -v MIN_START=2048 -v action=filldisk -v target=/dev/sda -v sizePos=146800640 -v diskSize=146800640 -v fixedList=1:5 in1.txt

Based on the most current this is what I see:

./procsfdisk.awk -v SECTOR_SIZE=512 -v CHUNK_SIZE=512 -v MIN_START=2048 -v action=filldisk -v target=/dev/sda -v sizePos=146800640 -v diskSize=146800640 -v fixedList=1:5 in1.txt
# Partition table is consistent.
label: dos
label-id: 0x77265efa
device: /dev/sda
unit: sectors

/dev/sda1 : start=        2048, size=     3940352, type=82, bootable
/dev/sda2 : start=     3944446, size=    71485442, type=5
/dev/sda5 : start=     3944448, size=     3940352, type=83
/dev/sda6 : start=     7884800, size=    67425792, type=83

[Sebastian Roth]

@Tom-Elliott Sorry, didn’t see that you pushed a new version. Confirmed that my error posted does not apply anymore! Looks good to me now.

Still I think about creating a testcases suite for the procsfdisk.awk script. So we could always check a lot of different layouts against the most current script to see if everything still works.

[Tom Elliott]

@sebastian-roth I agree, but data is hard to come by, and I don’t think we can come up with every case potential.