Inode exhaustion - the filesystem that says no · Linux

The disk is not full: df -h shows free space. But the filesystem says No space left on device. The cause is inode exhaustion: the filesystem has no free inodes to allocate to new files.

What an inode is

An inode is a metadata structure that describes a file: type, permissions, owner, timestamps, size, and pointers to the data blocks. Every file on an ext2/3/4 or XFS filesystem consumes one inode. The number of inodes is fixed at filesystem creation time.

How inode exhaustion happens

Read-only / Safedf -i— df -i reports inode usage. 100% used with 20 free means new files cannot be created. The filesystem has space (df -h shows plenty) but no inodes.

$ df -i /var

Filesystem     Inodes  IUsed   IFree IUse% Mounted on
/dev/sda2      5000000 4999980      20  100% /var

Where inodes go

Read-only / Safefile count— Counts files per mount. /var has 45000+ files - usually a mix of logs, caches, package state, and the like. Each of these consumes an inode.

$ for d in /var /tmp /home; do echo -n "$d: "; find $d -xdev | wc -l; done

/var: 45231
/tmp: 1203
/home: 87

Read-only / Safeinode hot spots— Find inode hot spots. /var/cache is often a culprit for inode exhaustion due to apt/dnf cache and other tooling. /var/spool/postfix is another common source of large file counts (one per message).

$ for d in /var/cache /var/log /var/spool; do echo -n "$d: "; find $d -xdev -type f | wc -l; done

/var/cache: 12345
/var/log: 234
/var/spool: 4567

Diagnosing and resolving

Run df -i to confirm inode exhaustion (IFree near 0)
Find the mountpoint consuming the most inodes: df -i per mount
Find directories with many files: find /mount -xdev -type f | wc -l per top-level dir
Identify the consumer: usually /var/cache, /var/spool/postfix/maildrop, /var/log/journal
Delete stale entries: apt clean, rotate logs, drop mail
Verify recovery: df -i
Long-term: reformat with more inodes via mkfs -N or -i, or use a filesystem with dynamic inode allocation (XFS)

XFS has dynamic inode allocation

XFS allocates inodes dynamically rather than fixing the count at mkfs time, so it does not hit the wall ext4 does. That is one of the reasons XFS is the default for large data filesystems.

It is not, however, unlimited. mkfs.xfs -i maxpct= caps the percentage of the filesystem that inodes may occupy, and the default is size-dependent: 25% under 1 TB, 5% under 50 TB, 1% above that. Hit that ceiling and allocations fail with ENOSPC while df still shows free space — the same confusing symptom as ext4 inode exhaustion, reached a different way. Check it before concluding XFS cannot run out:

xfs_info /data | grep -i imaxpct       # the cap, as a percentage
df -i /data                            # IUsed against IFree

maxpct can be raised after the fact with xfs_growfs -m, which is the one advantage over ext4 here.

For ext4 the inode count is fixed at format time and cannot be changed afterwards, so the choice has to be made up front. Note which flag does what — they are easy to swap and mke2fs accepts both:

Flag	Takes	Use
`-i bytes-per-inode`	A ratio	`-i 4096` makes one inode per 4 KiB of space, instead of the 16384 default. This is the option you want for many small files
`-N number-of-inodes`	An absolute count	`-N 2000000` creates exactly two million inodes, regardless of size

man 8 mke2fs on -i: “Be warned that it is not possible to change this ratio on a file system after it is created.” That is the whole reason to get it right at mkfs time — the recovery otherwise is a reformat and a restore.

For workloads that are mostly small files (mail spools, log archives, build caches), XFS is the better default.