nfs-storage – skill DevOps & Hạ tầng cho agent AI

NFS Storage

Configure NFS servers and clients for network file sharing across Linux systems. Covers NFSv4 server setup, export options, client mounting, autofs for on-demand mounts, Kerberos security, performance tuning, and Kubernetes integration.

When to Use

• Sharing directories between multiple Linux servers (web farms, build clusters)
• Providing shared storage for containerized workloads (Kubernetes ReadWriteMany)
• Centralizing home directories or application data across a fleet
• Setting up a development environment with shared project files
• Migrating from local storage to network-attached storage

Prerequisites

• NFS server: nfs-kernel-server (Debian/Ubuntu) or nfs-utils (RHEL/CentOS)
• NFS client: nfs-common (Debian/Ubuntu) or nfs-utils (RHEL/CentOS)
• Network connectivity between server and clients (TCP/UDP 2049 for NFSv4)
• Firewall rules allowing NFS traffic
• For NFSv4 Kerberos: krb5-user and a functioning KDC

NFS Server Setup

Installation

# Debian / Ubuntu
apt update && apt install -y nfs-kernel-server

# RHEL / CentOS
dnf install -y nfs-utils

# Enable and start the NFS server
systemctl enable --now nfs-server

# Verify NFS is running
systemctl status nfs-server
rpcinfo -p | grep nfs

Export Configuration (/etc/exports)

These are export-file entries, not shell commands. Limit clients and exports explicitly; root_squash is the safer baseline, not a substitute for client trust or authentication.

# /etc/exports
# Syntax: <directory> <client-spec>(options)

# Share /data to a specific subnet with read-write access
/data 10.0.0.0/24(rw,sync,no_subtree_check,root_squash)

# Share /shared read-only to everyone
/shared *(ro,sync,no_subtree_check)

# Share /home to specific hosts
/home server01.example.com(rw,sync,no_subtree_check)
/home server02.example.com(rw,sync,no_subtree_check)

# Share /var/nfs/projects with root squash (default, map root to nobody)
/var/nfs/projects 10.0.0.0/24(rw,sync,no_subtree_check,root_squash)

# NFSv4 pseudo-root export (recommended for NFSv4)
/srv/nfs 10.0.0.0/24(rw,sync,fsid=0,crossmnt,no_subtree_check)
/srv/nfs/data 10.0.0.0/24(rw,sync,no_subtree_check,root_squash)
/srv/nfs/shared 10.0.0.0/24(ro,sync,no_subtree_check)

Export Options Explained

| Option | Description |

|---|---|

| rw | Read-write access |

| ro | Read-only access |

| sync | Write data to disk before replying (safe, slower) |

| async | Reply before data is written to disk (fast, risk of corruption) |

| no_subtree_check | Disable subtree checking (improves reliability) |

| root_squash | Map remote root (UID 0) to nobody (default, more secure) |

| no_root_squash | Allow remote root to act as root on the server (use cautiously) |

| all_squash | Map all remote UIDs/GIDs to nobody |

| anonuid=1000 | Map anonymous users to a specific UID |

| anongid=1000 | Map anonymous groups to a specific GID |

| crossmnt | Allow clients to traverse into sub-mounts |

| fsid=0 | Mark as the NFSv4 pseudo-root |

Applying Export Changes

# Apply changes to exports (no server restart needed)
exportfs -ra

# Show current exports
exportfs -v

# Export a new directory on the fly (temporary, not persistent)
exportfs -o rw,sync,no_subtree_check 10.0.0.0/24:/tmp/share

# Unexport a directory
exportfs -u 10.0.0.0/24:/tmp/share

Server Firewall Configuration

# UFW (Ubuntu)
ufw allow from 10.0.0.0/24 to any port nfs
ufw allow from 10.0.0.0/24 to any port 111   # rpcbind (NFSv3)

# firewalld (RHEL/CentOS)
firewall-cmd --permanent --add-service=nfs
firewall-cmd --permanent --add-service=rpc-bind
firewall-cmd --permanent --add-service=mountd
firewall-cmd --reload

# NFSv4 only needs TCP 2049 (no rpcbind or mountd)
firewall-cmd --permanent --add-port=2049/tcp
firewall-cmd --reload

NFS Client Configuration

Manual Mounting

# Install NFS client
apt install -y nfs-common        # Debian/Ubuntu
dnf install -y nfs-utils         # RHEL/CentOS

# Discover exports from the server
showmount -e nfs-server.example.com

# Mount an NFS share manually
mkdir -p /mnt/data
mount -t nfs nfs-server.example.com:/data /mnt/data

# Mount with specific NFS version and options
mount -t nfs -o vers=4.2,tcp,hard,intr nfs-server.example.com:/data /mnt/data

# Verify the mount
mount | grep nfs
df -hT /mnt/data

# Unmount
umount /mnt/data

Persistent Mounts via /etc/fstab

These entries belong in fstab, not a shell.

# /etc/fstab entries for NFS

# Basic NFSv4 mount
nfs-server.example.com:/data  /mnt/data  nfs4  defaults,_netdev  0  0

# Mount with performance and reliability options
nfs-server.example.com:/data  /mnt/data  nfs4  hard,intr,rsize=1048576,wsize=1048576,timeo=600,retrans=3,_netdev  0  0

# Read-only mount
nfs-server.example.com:/shared  /mnt/shared  nfs4  ro,_netdev  0  0

# NFS identity mapping is configured through consistent server/client identities
# or supported NFSv4 idmapping/Kerberos, not generic uid=/gid= mount options.

# Mount all fstab entries
mount -a

# Test fstab entry without actually mounting
mount --fake -a -v

Mount Options Explained

| Option | Description |

|---|---|

| hard | Retry NFS requests indefinitely (recommended for data integrity) |

| soft | Return error after retrans retries (risk of data corruption) |

| intr | Historical option, ignored by modern Linux kernels; do not rely on it for interruption |

| rsize=1048576 | Read buffer size in bytes (1 MB, max for NFSv4) |

| wsize=1048576 | Write buffer size in bytes (1 MB) |

| timeo=600 | Timeout in tenths of a second (60 seconds) |

| retrans=3 | Number of retries before error (soft) or message (hard) |

| _netdev | Wait for network before mounting (critical for boot) |

| noatime | Do not update access time (improves performance) |

| nconnect=8 | Use multiple TCP connections (kernel 5.3+, improves throughput) |

Autofs (On-Demand Mounting)

# Install autofs
apt install -y autofs        # Debian/Ubuntu
dnf install -y autofs        # RHEL/CentOS

# Configure the master map
# /etc/auto.master or /etc/auto.master.d/nfs.autofs
/mnt/nfs  /etc/auto.nfs  --timeout=300

/etc/auto.nfs

# Format: mount-point  options  location
# Mounts will appear under /mnt/nfs/<mount-point>

data    -rw,hard,intr,rsize=1048576,wsize=1048576    nfs-server.example.com:/data
shared  -ro,hard,intr                                  nfs-server.example.com:/shared
home    -rw,hard,intr                                  nfs-server.example.com:/home/&

# Wildcard: mount any subdirectory from the server automatically
# /etc/auto.master entry:  /mnt/nfs  /etc/auto.nfs
*       -rw,hard,intr    nfs-server.example.com:/srv/nfs/&

# Enable and start autofs
systemctl enable --now autofs

# Test: simply cd into the mount point and it appears
ls /mnt/nfs/data    # Triggers auto-mount
# The share unmounts automatically after the timeout (300 seconds idle)

# Check autofs status
systemctl status autofs
automount -v    # Verbose debugging mode (foreground)

Before mutable exports, firewall/fstab changes, remounts or benchmarks, confirm host/path/client scope, preserve config and validate access/recovery. The dd/fio examples create/delete data and generate load: use an empty dedicated scratch export and unique file names, verify capacity, never run against user data. Kerberos is mentioned but not fully configured here; consult target distro guidance. See [Linux NFS mount semantics](https://man7.org/linux/man-pages/man5/nfs.5.html).

Performance Tuning

Server-Side Tuning

# Increase the number of NFS daemon threads (default is 8)
# /etc/default/nfs-kernel-server (Debian) or /etc/sysconfig/nfs (RHEL)
RPCNFSDCOUNT=32

# Or set at runtime
echo 32 > /proc/fs/nfsd/threads

# Restart NFS to apply
systemctl restart nfs-server

# Tune NFS server read/write sizes in sysctl
# These are auto-negotiated but can be adjusted
echo 1048576 > /proc/fs/nfsd/max_block_size

# Kernel network buffer tuning (see performance-tuning skill)
sysctl -w net.core.rmem_max=134217728
sysctl -w net.core.wmem_max=134217728

Client-Side Tuning

# Use large read/write buffer sizes in mount options
mount -t nfs4 -o rsize=1048576,wsize=1048576,noatime nfs-server:/data /mnt/data

# Use multiple TCP connections (Linux kernel 5.3+)
mount -t nfs4 -o nconnect=8 nfs-server:/data /mnt/data

# Check current mount options and NFS statistics
nfsstat -c                    # Client NFS statistics
nfsstat -s                    # Server NFS statistics
mountstats /mnt/data          # Detailed per-mount stats

# Test NFS throughput with dd
dd if=/dev/zero of=/mnt/data/testfile bs=1M count=1024 oflag=direct
dd if=/mnt/data/testfile of=/dev/null bs=1M iflag=direct
rm /mnt/data/testfile

# Test with fio for more realistic workloads
fio --name=nfs-test --directory=/mnt/data --ioengine=libaio --direct=1 \
  --rw=randrw --bs=4k --numjobs

… (nội dung đầy đủ có trong gói tải về)