5% more for free

This week I learned something new about Extended File System. On Thursday I received an email alert from administrator that one’s machine belongs to my team exceeded 90% of disk usage. It happened us first time and I was a little confused. Since a few months we are working on a new machine, moreover after migration we have much more disk space. On Friday went on production a new version of our software which extracts core dumps, checks backtracks and creates or update reports in issue tracking system. I decided that an investigation should be performed by the developer responsible for last upgrade. It wasn’t hard to check what is wrong, just login into that machine and review sizes of all directories. One hour later I had a report that it was an ordinary issue – we don’t clean up cache and it’s full of old data, but there was something more. The developer mentioned that we lose somewhere 5% of whole free space. It sounds strange – I said. I mentioned that Extended File System, ext4 in this case, like each file system takes part of a disk to store metadata like i-nodes. But, according to developer it was something different. He showed me calculations – sizes of directories, free space – and I had to admit that he was right.

After the stand-up meeting he said that he found out what it going on. When you create a new partition by default 5% of it’s reserved for system – more precisely – 5% of blocks. Detailed description comes from man for tune2fs:

-m reserved-blocks-percentage
       Set the percentage of the filesystem which may only be allocated by privileged processes.
       Reserving some number of filesystem blocks for  use  by  privileged  processes  is  done to
       avoid filesystem fragmentation, and to allow system daemons, such as syslogd(8), to continue
       to function correctly after non-privileged processes are prevented from writing to the filesystem.
       Normally, the default percentage of reserved blocks is 5%.

You may say that 5% it’s small part of disk and we shouldn’t care about it. As always everything depends on usage. If it’s a one big partition for everything, system and data, like at work, then I agree – we decided to not change it. But, when you have many partitions, dedicated only for data or casual usb stick then you wish use whole available space. Small example from my usb backup disk, it has 1TB:

dirdival@discworld:~$ sudo tune2fs -l /dev/sdd1
...
Filesystem OS type:       Linux
Inode count:              61054976
Block count:              244190208
Reserved block count:     12209510
Free blocks:              63679274
Free inodes:              61054956
First block:              0
Block size:               4096
...

Quick calculation shows that I’m wasting 47693MB, more or less 46GB.

>>> (12209510 * 4096) / 1024**2
47693
>>> 47693 / 1024
46

In this case scale has a meaning, almost 47GB more for free makes a different for a backup disk. From that purpose I decided to change that value into 0, it’s easy to obtain:

dirdival@discworld:~$ sudo tune2fs -m 0  /dev/sdd1

After that I have:

dirdival@discworld:~$ sudo tune2fs -l /dev/sdd1
...
Filesystem OS type:       Linux
Inode count:              61054976
Block count:              244190208
Reserved block count:     0
Free blocks:              63679274
Free inodes:              61054956
First block:              0
...

And small proof that it was good idea. Before change:

dirdival@discworld:~$ df -h
...
/dev/sdd1                  917G  674G  197G  78% /media/cdrom

and after:

dirdival@discworld:~$ df -h
...
/dev/sdd1                  917G  674G  243G  74% /media/cdrom

Of course I made change on unmounted partition.

---

Thank you Marcin for your deeper investigation, I really appreciate it.