When working with ext3 filesystems on Linux servers, you might encounter the warning:

EXT3-fs warning (device dm-3): ext3_dx_add_entry: Directory index full!

This typically occurs when the directory hash index (HTREE) reaches its maximum capacity. The ext3 filesystem uses a hashed B-tree (HTREE) structure to speed up directory lookups, but each directory node has a fixed limit of around 10,000-20,000 entries.

First, verify your current inode usage:

df -i
Filesystem            Inodes   IUsed   IFree IUse% Mounted on
/dev/mapper/ddf1_p3  77037568 9996012 67041556   13% /

Then check filesystem features with:

tune2fs -l /dev/mapper/ddf1_p3 | grep features
Filesystem features: has_journal ext_attr resize_inode dir_index filetype needs_recovery sparse_super large_file

Immediate workaround:

# Temporarily disable directory indexing:
debugfs -w /dev/mapper/ddf1_p3
debugfs: features ^dir_index
debugfs: quit

Permanent solution:

# 1. Backup the directory
rsync -a /problem/directory/ /backup/location/

# 2. Unmount the filesystem
umount /mountpoint

# 3. Run e2fsck with HTREE rebuild
e2fsck -fD /dev/mapper/ddf1_p3

# 4. Remount
mount /mountpoint

For web servers serving static assets, consider these architectural improvements:

# Example Nginx configuration to split directories
location ~ ^/assets/([a-z0-9])/([a-z0-9])/(.*)$ {
    alias /data/assets/$1/$2/$3;
}

Or implement a sharding script for existing directories:

#!/bin/bash
# Directory sharding script
for file in /data/www/lighttpd/images/*; do
    hash=$(echo "$file" | md5sum | cut -c1-2)
    mkdir -p "/data/www/lighttpd/images/$hash"
    mv "$file" "/data/www/lighttpd/images/$hash/"
done

If the issue persists after HTREE rebuild, consider:

1. Upgrading to ext4 (which has improved HTREE implementation)
2. Implementing XFS for directories with millions of files
3. Checking for filesystem corruption with fsck -fn /dev/device

---

During routine FTP operations via proftpd, kernel logs started showing repetitive warnings:

EXT3-fs warning (device dm-3): ext3_dx_add_entry: Directory index full!
Sep 16 15:30:34 xx last message repeated 489 times

This occurred while serving static assets through lighttpd 1.4.28-1 on CentOS 5.3, despite the filesystem showing plenty of available inodes (13% usage).

The ext3 filesystem uses two directory indexing methods:

1. Classic linear directory (old-style)
2. Hash-based directory index (dx)

The dx index has a fixed-size hash table (by default 2 levels deep) that can store about:

- First level: ~500 entries
- Second level: ~65,000 entries

The error occurs when:

1. The directory uses dx indexing (dir_index feature enabled)
2. The directory contains more than ~65k files
3. The hash table becomes full and can't grow further

Key verification commands:

# Check filesystem features
tune2fs -l /dev/mapper/ddf1_p3 | grep features

# Count actual files in problematic directory
find /data/www/lighttpd -xdev -type f | wc -l

Immediate Workaround

Disable directory indexing temporarily:

# For existing directories:
debugfs -w /dev/mapper/ddf1_p3
rm_dir_index /data/www/lighttpd/problem_directory
quit

Or recreate the directory:

mkdir /data/www/lighttpd/new_dir
rsync -a /data/www/lighttpd/problem_directory/ /data/www/lighttpd/new_dir/
rm -rf /data/www/lighttpd/problem_directory
mv /data/www/lighttpd/new_dir /data/www/lighttpd/problem_directory

Long-term Solutions

Option 1: Directory restructuring

# Example sharding script
for file in /data/www/lighttpd/large_dir/*
do
  prefix=$(echo $file | md5sum | cut -c1-2)
  mkdir -p "/data/www/lighttpd/shards/$prefix"
  mv "$file" "/data/www/lighttpd/shards/$prefix/"
done

Option 2: Filesystem conversion

# Convert to ext4 (supports larger dx indexes)
tune2fs -O dir_index,extents,uninit_bg /dev/mapper/ddf1_p3
fsck -pf /dev/mapper/ddf1_p3

For lighttpd configurations:

# In lighttpd.conf
server.follow-symlink = "enable"
server.upload-dirs = ( "/var/tmp" )

# Enable mmap for static files
server.network-backend = "writev"

Filesystem monitoring script example:

#!/bin/bash
THRESHOLD=50000
DIRS=$(find /data/www/lighttpd -type d)

for dir in $DIRS; do
  count=$(find "$dir" -maxdepth 1 -type f | wc -l)
  if [ $count -gt $THRESHOLD ]; then
    logger -t DIRWARN "Large directory detected: $dir ($count files)"
  fi
done