Utilities for Amazon Elastic File System (EFS)
The efs-utils
package has been verified against the following Linux distributions:
Distribution | Package Type | init System |
---|---|---|
Amazon Linux 2017.09 | rpm |
upstart |
Amazon Linux 2 | rpm |
systemd |
CentOS 7 | rpm |
systemd |
CentOS 8 | rpm |
systemd |
RHEL 7 | rpm |
systemd |
RHEL 8 | rpm |
systemd |
Fedora 28 | rpm |
systemd |
Fedora 29 | rpm |
systemd |
Fedora 30 | rpm |
systemd |
Fedora 31 | rpm |
systemd |
Fedora 32 | rpm |
systemd |
Debian 9 | deb |
systemd |
Debian 10 | deb |
systemd |
Ubuntu 16.04 | deb |
systemd |
Ubuntu 18.04 | deb |
systemd |
Ubuntu 20.04 | deb |
systemd |
OpenSUSE Leap | rpm |
systemd |
OpenSUSE Tumbleweed | rpm |
systemd |
Oracle8 | rpm |
systemd |
SLES 12 | rpm |
systemd |
SLES 15 | rpm |
systemd |
The efs-utils
package has been verified against the following MacOS distributions:
Distribution | init System |
---|---|
MacOS Big Sur | launchd |
MacOS Monterey | launchd |
MacOS Ventura | launchd |
- Prerequisites
- Optional
- Installation
- Usage
- Troubleshooting
- Upgrading stunnel for RHEL/CentOS
- Upgrading stunnel for SLES12
- Upgrading stunnel for MacOS
- Install botocore
- Upgrade botocore
- Enable mount success/failure notification via CloudWatch log
- Optimize readahead max window size on Linux 5.4+
- Using botocore to retrieve mount target ip address when dns name cannot be resolved
- The way to access instance metadata
- Use the assumed profile credentials for IAM
- Enabling FIPS Mode
- License Summary
nfs-utils
(RHEL/CentOS/Amazon Linux/Fedora) ornfs-common
(Debian/Ubuntu)- OpenSSL 1.0.2+
- Python 3.4+
stunnel
4.56+
botocore
1.12.0+
For those using Amazon Linux or Amazon Linux 2, the easiest way to install efs-utils
is from Amazon's repositories:
$ sudo yum -y install amazon-efs-utils
You can now use AWS Systems Manage Distributor to automatically install or update amazon-efs-utils
.
Please refer to Using AWS Systems Manager to automatically install or update Amazon EFS clients for more guidance.
The following are prerequisites for using AWS Systems Manager Distributor to install or update amazon-efs-utils
:
-
AWS Systems Manager agent is installed on the distribution (For
Amazon Linux
andUbuntu
, AWS Systems Manager agent is pre-installed, for other distributions, please refer to install AWS Systems Manager agent on Linux EC2 instance for more guidance.) -
Instance is attached with IAM role with AWS managed policy
AmazonElasticFileSystemsUtils
, this policy will enable your instance to be managed by AWS Systems Manager agent, also it contains permissions to support specific features.
Other distributions require building the package from source and installing it.
- To build and install an RPM:
If the distribution is not OpenSUSE or SLES
$ sudo yum -y install git rpm-build make
$ git clone https://github.com/aws/efs-utils
$ cd efs-utils
$ make rpm
$ sudo yum -y install build/amazon-efs-utils*rpm
Otherwise
$ sudo zypper refresh
$ sudo zypper install -y git rpm-build make
$ git clone https://github.com/aws/efs-utils
$ cd efs-utils
$ make rpm
$ sudo zypper --no-gpg-checks install -y build/amazon-efs-utils*rpm
On OpenSUSE, if you see error like File './suse/noarch/bash-completion-2.11-2.1.noarch.rpm' not found on medium 'http://download.opensuse.org/tumbleweed/repo/oss/'
during installation of git
, run the following commands to re-add repo OSS and NON-OSS, then run the install script above again.
sudo zypper ar -f -n OSS http://download.opensuse.org/tumbleweed/repo/oss/ OSS
sudo zypper ar -f -n NON-OSS http://download.opensuse.org/tumbleweed/repo/non-oss/ NON-OSS
sudo zypper refresh
- To build and install a Debian package:
$ sudo apt-get update
$ sudo apt-get -y install git binutils
$ git clone https://github.com/aws/efs-utils
$ cd efs-utils
$ ./build-deb.sh
$ sudo apt-get -y install ./build/amazon-efs-utils*deb
For EC2 Mac instances running macOS Big Sur, macOS Monterey and macOS Ventura, you can install amazon-efs-utils from the homebrew-aws respository. Note that this will ONLY work on EC2 instances running macOS Big Sur, macOS Monterey and macOS Ventura, not local Mac computers.
brew install amazon-efs-utils
This will install amazon-efs-utils on your EC2 Mac Instance running macOS Big Sur, macOS Monterey and macOS Ventura in the directory /usr/local/Cellar/amazon-efs-utils
.
Follow the instructions in caveats when using efs-utils on EC2 Mac instance for the first time. To check the package caveats run below command
brew info amazon-efs-utils
- Set up a virtualenv for efs-utils
$ virtualenv ~/.envs/efs-utils
$ source ~/.envs/efs-utils/bin/activate
$ pip install -r requirements.txt
- Run tests
$ make test
efs-utils
includes a mount helper utility to simplify mounting and using EFS file systems.
To mount with the recommended default options, simply run:
$ sudo mount -t efs file-system-id efs-mount-point/
To mount file system to a specific mount target of the file system, run:
$ sudo mount -t efs -o mounttargetip=mount-target-ip-address file-system-id efs-mount-point/
To mount file system within a given network namespace, run:
$ sudo mount -t efs -o netns=netns-path file-system-id efs-mount-point/
To mount file system to the mount target in specific availability zone (e.g. us-east-1a), run:
$ sudo mount -t efs -o az=az-name file-system-id efs-mount-point/
To mount over TLS, simply add the tls
option:
$ sudo mount -t efs -o tls file-system-id efs-mount-point/
To authenticate with EFS using the system’s IAM identity, add the iam
option. This option requires the tls
option.
$ sudo mount -t efs -o tls,iam file-system-id efs-mount-point/
To mount using an access point, use the accesspoint=
option. This option requires the tls
option.
The access point must be in the "available" state before it can be used to mount EFS.
$ sudo mount -t efs -o tls,accesspoint=access-point-id file-system-id efs-mount-point/
To not use the DNS name in the mount command, and force the use of the resolved IP instead. This option requires that fall_back_to_mount_target_ip_address_enabled = true
is set in efs-utils.conf under the [mount]
section.
$ sudo mount -t efs -o forceuseip file-system-id efs-mount-point/
To mount your file system automatically with any of the options above, you can add entries to /efs/fstab
like:
file-system-id efs-mount-point efs _netdev,tls,iam,accesspoint=access-point-id 0 0
For more information on mounting with the mount helper, see the manual page:
man mount.efs
or refer to the documentation.
For EC2 instances using Mac distribution, the recommended default options will perform a tls mount:
$ sudo mount -t efs file-system-id efs-mount-point/
or
$ sudo mount -t efs -o tls file-system-id efs-mount-point/
To mount without TLS, simply add the notls
option:
$ sudo mount -t efs -o notls file-system-id efs-mount-point/
efs-utils
contains a watchdog process to monitor the health of TLS mounts. This process is managed by either upstart
or systemd
depending on your Linux distribution and launchd
on Mac distribution, and is started automatically the first time an EFS file system is mounted over TLS.
If you run into a problem with efs-utils, please open an issue in this repository. We can more easily
assist you if relevant logs are provided. You can find the log file at /var/log/amazon/efs/mount.log
.
Often times, enabling debug level logging can help us find problems more easily. To do this, run
sed -i '/logging_level = INFO/s//logging_level = DEBUG/g' /etc/amazon/efs/efs-utils.conf
.
You can also enable stunnel debug logs with
sed -i '/stunnel_debug_enabled = false/s//stunnel_debug_enabled = true/g' /etc/amazon/efs/efs-utils.conf
.
Make sure to perform the failed mount again after running the prior commands before pulling the logs.
By default, when using the EFS mount helper with TLS, it enforces certificate hostname checking. The EFS mount helper uses the stunnel
program for its TLS functionality. Please note that some versions of Linux do not include a version of stunnel
that supports TLS features by default. When using such a Linux version, mounting an EFS file system using TLS will fail.
Once you’ve installed the amazon-efs-utils
package, to upgrade your system’s version of stunnel
, see Upgrading Stunnel.
Run the following commands and follow the output hint of zypper package manager to upgrade the stunnel on your SLES12 instance
sudo zypper addrepo https://download.opensuse.org/repositories/security:Stunnel/SLE_12_SP5/security:Stunnel.repo
sudo zypper refresh
sudo zypper install -y stunnel
The installation installs latest stunnel available in brew repository. You can also upgrade the version of stunnel on your instance using the command below:
brew upgrade stunnel
efs-utils
uses botocore to interact with other AWS services. Please note the package type from the above table and install
botocore based on that info. If botocore is already installed and does not meet the minimum required version,
you can upgrade the botocore by following the upgrade botocore section.
- Download the
get-pip.py
script
sudo yum -y install wget
if [[ "$(python3 -V 2>&1)" =~ ^(Python 3.6.*) ]]; then
sudo wget https://bootstrap.pypa.io/pip/3.6/get-pip.py -O /tmp/get-pip.py
elif [[ "$(python3 -V 2>&1)" =~ ^(Python 3.5.*) ]]; then
sudo wget https://bootstrap.pypa.io/pip/3.5/get-pip.py -O /tmp/get-pip.py
elif [[ "$(python3 -V 2>&1)" =~ ^(Python 3.4.*) ]]; then
sudo wget https://bootstrap.pypa.io/pip/3.4/get-pip.py -O /tmp/get-pip.py
else
sudo wget https://bootstrap.pypa.io/get-pip.py -O /tmp/get-pip.py
fi
sudo apt-get update
sudo apt-get -y install wget
if echo $(python3 -V 2>&1) | grep -e "Python 3.6"; then
sudo wget https://bootstrap.pypa.io/pip/3.6/get-pip.py -O /tmp/get-pip.py
elif echo $(python3 -V 2>&1) | grep -e "Python 3.5"; then
sudo wget https://bootstrap.pypa.io/pip/3.5/get-pip.py -O /tmp/get-pip.py
elif echo $(python3 -V 2>&1) | grep -e "Python 3.4"; then
sudo wget https://bootstrap.pypa.io/pip/3.4/get-pip.py -O /tmp/get-pip.py
else
sudo apt-get -y install python3-distutils
sudo wget https://bootstrap.pypa.io/get-pip.py -O /tmp/get-pip.py
fi
- To install botocore on RPM
sudo python3 /tmp/get-pip.py
sudo pip3 install botocore || sudo /usr/local/bin/pip3 install botocore
- To install botocore on DEB
sudo python3 /tmp/get-pip.py
sudo pip3 install botocore || sudo /usr/local/bin/pip3 install botocore
sudo python3 /tmp/get-pip.py
sudo pip3 install --target /usr/lib/python3/dist-packages botocore || sudo /usr/local/bin/pip3 install --target /usr/lib/python3/dist-packages botocore
sudo pip3 install botocore
Pass --upgrade
to the corresponding installation scripts above based on system platform and distribution
sudo pip3 install botocore --upgrade
efs-utils
now support publishing mount success/failure logs to CloudWatch log. By default, this feature is disabled. There are three
steps you must follow to enable and use this feature:
Follow install botocore section
sudo sed -i -e '/\[cloudwatch-log\]/{N;s/# enabled = true/enabled = true/}' /etc/amazon/efs/efs-utils.conf
- For MacOS:
EFS_UTILS_VERSION=<e.g. 1.34.5>
sudo sed -i -e '/\[cloudwatch-log\]/{N;s/# enabled = true/enabled = true/;}' /usr/local/Cellar/amazon-efs-utils/${EFS_UTILS_VERSION}/libexec/etc/amazon/efs/efs-utils.conf
- For Mac2 instance:
EFS_UTILS_VERSION=<e.g. 1.34.5>
sudo sed -i -e '/\[cloudwatch-log\]/{N;s/# enabled = true/enabled = true/;}' /opt/homebrew/Cellar/amazon-efs-utils/${EFS_UTILS_VERSION}/libexec/etc/amazon/efs/efs-utils.conf
You can also configure CloudWatch log group name and log retention days in the config file.
If you want to have separate log groups in Cloudwatch for every mounted file system, add /{fs_id}
to the end of the log_group_name
field in efs-utils.conf
file. For example, the log_group_name
in efs-utils.conf
file would look something like:
[cloudwatch-log]
log_group_name = /aws/efs/utils/{fs_id}
Attach AWS managed policy AmazonElasticFileSystemsUtils
to the iam role you attached to the instance, or the aws credentials
configured on your instance.
After completing the three prerequisite steps, you will be able to see mount status notifications in CloudWatch Logs.
A change in the Linux kernel 5.4+ results a throughput regression on NFS client. With patch, starting from 5.4.*, Kernels containing this patch now set the default read_ahead_kb size to 128 KB instead of the previous 15 MB. This read_ahead_kb is used by the Linux kernel to optimize performance on NFS read requests by defining the maximum amount of data an NFS client can pre-fetch in a read call. With the reduced value, an NFS client has to make more read calls to the file system, resulting in reduced performance.
To avoid above throughput regression, efs-utils will modify read_ahead_kb to 15 * rsize (could be configured via mount option, 1MB by default) after mount success on Linux 5.4+. (not support on MacOS)
This optimization will be enabled by default. To disable this optimization:
sed -i "s/optimize_readahead = false/optimize_readahead = true/" /etc/amazon/efs/efs-utils.conf
To re-enable this optimization
sed -i "s/optimize_readahead = true/optimize_readahead = false/" /etc/amazon/efs/efs-utils.conf
You can mount file system with a given rsize, run:
$ sudo mount -t efs -o rsize=rsize-value-in-bytes file-system-id efs-mount-point/
You can also manually chose a value of read_ahead_kb to optimize read throughput on Linux 5.4+ after mount.
$ sudo bash -c "echo read-ahead-value-in-kb > /sys/class/bdi/0:$(stat -c '%d' efs-mount-point)/read_ahead_kb"
efs-utils
now supports using botocore to retrieve mount target ip address when dns name cannot be resolved, e.g.
when user is mounting a file system in another VPC. There are two prerequisites to use this feature:
Follow install botocore section
Allow the elasticfilesystem:DescribeMountTargets
and ec2:DescribeAvailabilityZones
action in your policy attached to
the iam role you attached to the instance, or the aws credentials configured on your instance. We recommend you attach
AWS managed policy AmazonElasticFileSystemsUtils
.
This feature will be enabled by default. To disable this feature:
sed -i "s/fall_back_to_mount_target_ip_address_enabled = true/fall_back_to_mount_target_ip_address_enabled = false/" /etc/amazon/efs/efs-utils.conf
If you decide that you do not want to use this feature, but need to mount a cross-VPC file system, you can use the mounttargetip option to do so, using the desired mount target ip address in the mount command.
efs-utils
by default uses IMDSv2, which is a session-oriented method used to access instance metadata. If you don't want to use
IMDSv2, you can disable the token fetching feature by running the following command:
sed -i "s/disable_fetch_ec2_metadata_token = false/disable_fetch_ec2_metadata_token = true/" /etc/amazon/efs/efs-utils.conf
To authenticate with EFS using the system’s IAM identity of an awsprofile, add the iam
option and pass the profile name to
awsprofile
option. These options require the tls
option.
$ sudo mount -t efs -o tls,iam,awsprofile=test-profile file-system-id efs-mount-point/
To configure the named profile, see the Named Profiles doc
and Support Config File Settings doc
for more details. If the credentials (e.g. aws_access_key_id) are not configured in /root/.aws/credentials
or /root/.aws/config
(note that the path prefix may vary based on the root path of sudo), efs-utils will use botocore to assume the named profile.
This will require botocore is pre-installed, please follow install botocore section to install botocore first.
Normally you will need to configure your profile IAM policy to make the assume works. For example, if you want to perform a cross-account mounting, suppose you have established vpc-peering-connections between your vpcs, next step you need to do is giving permission to account B so that it can assume a role in account A and then mount the file system that belongs to account A. You can see IAM doc for more details.
After the IAM identity is setup, you can configure your awsprofile credentials or config. You can refer to
sdk settings. For example you can define
the profile to use the credentials of profile default
to assume role in account A by defining the source_profile
.
# /root/.aws/credentials
[default]
aws_access_key_id = AKIAIOSFODNN7EXAMPLE
aws_secret_access_key_id =wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY
# /root/.aws/config
[default]
...
[profile test-profile]
role_arn = <role-arn-in-account-A>
source_profile = default
Or you can use the credentials from IAM role attached to instance to assume the named profile, e.g.
# /root/.aws/config
[profile test-profile]
role_arn = <role-arn-in-account-A>
credential_source = Ec2InstanceMetadata
You can use web identity to assume a role which has the permission to attach to the EFS filesystem. You need to have a valid JWT token and a role arn to assume. There are two ways you can leverage them:
- By setting environment variable the path to the file containing the JWT token in
AWS_WEB_IDENTITY_TOKEN_FILE
and by settingROLE_ARN
environment variable. The command below shows an example of to leverage it.
$ sudo mount -t efs -o tls,iam file-system-id efs-mount-point/
- By passing the JWT token file path and the role arn as parameters to the mount command. The command below shows an example of to leverage it.
$ sudo mount -t efs -o tls,iam,rolearn="ROLE_ARN",jwtpath="PATH/JWT_TOKEN_FILE" file-system-id efs-mount-point/
Efs-Utils is able to enter FIPS mode when mounting your file system. To enable FIPS you need to modify the EFS-Utils config file:
sed -i "s/fips_mode_enabled = false/fips_mode_enabled = true/" /etc/amazon/efs/efs-utils.conf
This will enable any potential API call from EFS-Utils to use FIPS endpoints and cause stunnel to enter FIPS mode
Note: FIPS mode requires that the installed version of OpenSSL is compiled with FIPS.
To verify that the installed version is compiled with FIPS, look for OpenSSL X.X.Xx-fips
in the stunnel -version
command output e.g.
$ stunnel -version
stunnel 4.56 on x86_64-koji-linux-gnu platform
Compiled/running with OpenSSL 1.0.2k-fips 26 Jan 2017
Threading:PTHREAD Sockets:POLL,IPv6 SSL:ENGINE,OCSP,FIPS Auth:LIBWRAP
For more information on how to configure OpenSSL with FIPS see the OpenSSL FIPS README.
This code is made available under the MIT license.