How to Mount an S3 Bucket on Linux and Windows: Mountpoint for Amazon S3, fstab, S3 Files, the EKS CSI Driver, rclone and Every Error

Logeshwaran
—

To mount an S3 bucket as a drive or folder, pick the tool by where you need it. On Linux (EC2 or your own servers), install Mountpoint for Amazon S3, a free, open-source client from AWS, and run mount-s3 my-bucket /mnt/my-bucket; it gives fast reads and lets you create new files, but you cannot edit an existing file in place, rename on normal buckets, or use symlinks, and deletes need --allow-delete. If your app needs a real file system (edits, renames, locks, shared writes), use Amazon S3 Files, the managed NFS file system for S3 buckets that went live in April 2026, mounted with mount -t s3files. On Windows, Mountpoint is not available: use rclone mount with WinFsp for a drive letter, or an S3 File Gateway for an SMB share. On EKS, use the Mountpoint CSI driver (not on Fargate). Mountpoint itself costs nothing; you pay the normal S3 request and storage charges. Below: install commands, mount options, fstab for mounting at boot, IAM permissions, caching, Kubernetes, Windows and macOS, s3fs compared, costs, and the errors you will hit.

Jake's reporting server ran a Python script that expected a folder of CSV files at /data/reports. The CSVs lived in an S3 bucket, so a cron job synced them down every hour, and every few weeks the disk filled up and the reports stopped. Ethan's suggestion was to stop copying: mount the bucket and let the script read the files straight from S3.

⚡ Quick Answer

• Linux or EC2: Mountpoint for Amazon S3, mount-s3 bucket /mnt/dir. Install and mount.

• Need edits, renames, locks: Amazon S3 Files (NFS). S3 Files.

• Windows drive letter: rclone + WinFsp, or S3 File Gateway. Windows.

• Mount at boot: one fstab line. fstab.

Which tool · What works · EKS · macOS · s3fs vs Mountpoint · Cost · Errors.

New to S3, or not sure whether a file system is what you need? The cards below explain S3, the three kinds of storage, and the permissions every mount depends on.

🧭 NEW HERE? READ THESE FIRST

New to S3 or IAM? These five pages give you the background this one builds on:

📌 Bookmark this; the which-tool table and the fstab line are the parts you will come back for.

Which tool to use to mount S3 on your system

S3 is object storage, not a disk. Every "mount" tool turns file operations into S3 requests, and each one makes different trade-offs. Start from where your code runs and what it does with files:

Where you need the filesBest toolWhy
Linux EC2 or on-premises Linux, mostly reading big filesMountpoint for Amazon S3Free, very fast reads, simple
Linux apps that edit files, rename, lock, or share writesAmazon S3 FilesManaged NFS on top of the bucket, synced both ways
EKS pods on EC2 nodesMountpoint CSI driver, or S3 FilesBucket as a persistent volume
ECS tasks on Fargate, Lambda functionsAmazon S3 FilesFargate cannot run FUSE clients; S3 Files volumes can
Windows PC or Windows Server, drive letterrclone mount + WinFspFree; Mountpoint is Linux only
Office users on an SMB shareS3 File GatewayA normal network share, files stored as S3 objects
Macrclone (with macFUSE, FUSE-T, or nfsmount)Mountpoint does not run on macOS
Older Linux scripts that need chmod or in-place edits on a budgets3fs-fuseMore POSIX features, slower

If you only need a copy of the files, a mount may be the wrong answer: aws s3 sync on a schedule is simpler. And if the data is really a shared working folder with no need to be S3 objects, a plain file system such as Amazon EFS fits better.

Jake: "So it isn't one tool. My server is Amazon Linux, the script only reads."

Ethan: "Then it's Mountpoint. Five minutes, and your disk stops filling up."

Install Mountpoint for Amazon S3 on Linux

Mountpoint is a single binary, mount-s3, built on FUSE. Packages exist for x86_64 and ARM64 (Graviton). Pick the steps for your distribution:

  1. Amazon Linux 2023 (recent releases): it is in the normal repository.
    sudo dnf install mount-s3
  2. Amazon Linux 2, RHEL, CentOS, Fedora: download the RPM for your architecture and install it.
    wget https://s3.amazonaws.com/mountpoint-s3-release/latest/x86_64/mount-s3.rpm
    sudo yum install ./mount-s3.rpm
    For Graviton, swap x86_64 for arm64 in the URL.
  3. Ubuntu and Debian: the same idea with the DEB package.
    wget https://s3.amazonaws.com/mountpoint-s3-release/latest/x86_64/mount-s3.deb
    sudo apt-get install ./mount-s3.deb
  4. Other distributions: install your distribution's fuse and libfuse2 packages, unpack mount-s3.tar.gz from the same location into /opt/aws/mountpoint-s3, and add its bin folder to your PATH.
  5. Check it: mount-s3 --version prints the version, for example mount-s3 1.21.0.

Each package also has a .asc signature file next to it, so if your security team wants proof the package was not tampered with, import the public key and run gpg --verify mount-s3.rpm.asc; you are looking for "Good signature".

Mount the bucket and give it permissions

Make an empty folder and mount the bucket onto it. Mountpoint runs in the background, so the command returns straight away:

mkdir ~/reports
mount-s3 shop-reports-bucket ~/reports
ls ~/reports
umount ~/reports

Keys become paths: an object called daily/2026-10-20.csv shows up as a file 2026-10-20.csv inside a folder daily. Useful options you will reach for:

  • --prefix daily/: mount only part of the bucket (the prefix must end with /).
  • --read-only: block every change; the safe choice for reporting and training jobs.
  • --allow-delete and --allow-overwrite: turn on deletes and full-file overwrites (off by default).
  • --region eu-west-2: set the Region yourself if detection fails.
  • --profile reports: use a named profile from your AWS config files.
  • --storage-class and --sse aws:kms: the storage class and encryption for new files you write.
  • -f (foreground) and --debug: keep it in the terminal and print detail while you troubleshoot.

Credentials. Mountpoint uses the normal AWS credential sources: an EC2 instance profile, an ECS task role, a role profile in ~/.aws/config, or keys in the credentials file or environment variables. IAM Identity Center (SSO) profiles are not a supported source, which catches out people who mount from their laptop session. On EC2, attach a role through an instance profile and skip keys entirely.

Permissions. A read-only mount needs s3:ListBucket on the bucket and s3:GetObject on its objects. A read-write mount adds the write actions:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "MountpointBucket",
      "Effect": "Allow",
      "Action": ["s3:ListBucket"],
      "Resource": ["arn:aws:s3:::shop-reports-bucket"]
    },
    {
      "Sid": "MountpointObjects",
      "Effect": "Allow",
      "Action": ["s3:GetObject", "s3:PutObject", "s3:AbortMultipartUpload", "s3:DeleteObject"],
      "Resource": ["arn:aws:s3:::shop-reports-bucket/*"]
    }
  ]
}

If the bucket uses SSE-KMS, the role also needs kms:Decrypt to read and kms:GenerateDataKey to write; without them you get access errors that look like S3 problems (see our S3 KMS access denied guide). Our bucket access guide covers bucket policies and cross-account mounts.

What works on a Mountpoint mount, and what does not

This is the part that decides whether Mountpoint suits you. It is built for high-throughput reading and for writing new files in one go, and it deliberately skips the parts of a normal file system that S3 cannot do cheaply:

OperationWorks?Notes
List folders, read files, seek while readingYesMany readers at once; the main strength
Create a new fileYesWritten start to finish; visible to others after you close it
Replace a whole existing fileWith --allow-overwriteThe program must open it with truncate (O_TRUNC)
Edit part of a file, random writesNoUse S3 Files or s3fs
Append to a fileOnly S3 Express One ZoneWith --incremental-upload
Delete a fileWith --allow-deleteDeleted in S3 immediately
Rename or move a fileOnly directory bucketsNot on general purpose buckets; folders cannot be renamed anywhere
mkdirPartlyLocal until a file is written inside it
Symlinks, hard links, chmod, chown, file locks, xattrsNoOwnership and modes come from mount options
Read Glacier Flexible Retrieval or Deep Archive objectsNoRestore them first; other storage classes read fine

Two practical results. First, editors such as vim and tools such as sed -i write a temporary file and rename it, so they fail on a Mountpoint mount of a normal bucket; edit locally and copy the file in. Second, if you turn on --allow-delete, turn on bucket versioning too, so an accidental rm -r can be undone; our restore deleted S3 objects guide shows how.

Consistency is good: new files and writes are visible immediately to readers on the same mount. If another client changes an object, a stat can show old details for about a second, and longer if you turn on metadata caching (below).

Jake: "The script writes one summary file at the end. Does that work?"

Ethan: "Yes, a new file written in one go is exactly what it likes. Rewriting it next day needs --allow-overwrite."

Mount the bucket at boot with /etc/fstab

Since version 1.18, Mountpoint can mount from /etc/fstab. The steps:

  1. Create the folder: sudo mkdir -p /data/reports.
  2. Add one line to /etc/fstab, with the bucket written as an s3:// URL:
    s3://shop-reports-bucket/daily/ /data/reports mount-s3 _netdev,nosuid,nodev,nofail,rw,allow-other 0 0
  3. Reload systemd and mount: sudo systemctl daemon-reload, then sudo mount -a.
  4. Check it: findmnt /data/reports or df -h /data/reports; if it failed, systemctl status data-reports.mount shows why.
  5. Reboot once to confirm it comes back by itself.

The options matter. _netdev, nosuid and nodev are required; Mountpoint refuses to start without them. nofail lets the server boot even if the mount fails, which is what you want for a network mount. Command-line flags go in the options field without the dashes, as key=value pairs: --uid 1001 --allow-delete becomes uid=1001,allow-delete. Use ro instead of rw for a read-only mount.

From fstab, Mountpoint runs as root and reads credentials from root's home folder, not yours. On EC2 that is no problem with an instance profile; on your own servers, put the credentials where root can read them.

For new EC2 instances, put the install and the fstab line in the launch template's user data:

#!/bin/bash -e
dnf install -y mount-s3
mkdir -p /data/reports
echo "s3://shop-reports-bucket/daily/ /data/reports mount-s3 _netdev,nosuid,nodev,nofail,rw,allow-other 0 0" >> /etc/fstab
systemctl daemon-reload
mount -a

Let other users and services read the mount

By default only the user who ran mount-s3 can see inside the folder; even root gets "Permission denied". This is the most common surprise when a web server or a service account needs the files. To share the mount:

  • Add the line user_allow_other to /etc/fuse.conf.
  • Mount with --allow-other (or allow-other in fstab).
  • Set the owner and modes everyone sees with --uid, --gid, --file-mode and --dir-mode. Files default to 0644 and folders to 0755, owned by the mounting user.

Those modes are only what Linux shows. What the mount can actually do is still decided by the IAM role, so a 0644 file on a read-only role is still read-only in S3.

Speed up repeated reads with caching

Without caching, every read goes to S3, which is fine for reading each file once. If you read the same files again and again (training data, reference files, a dashboard that reloads), turn on a cache:

  • Local cache: --cache /var/cache/mountpoint, optionally with --max-cache-size in MiB. Uses instance storage or an EBS volume. The cache is emptied when you mount and unmount.
  • Shared cache: --cache-xz my-cache--usw2-az1--x-s3 keeps small objects (up to 1 MB) in an S3 Express One Zone directory bucket that many instances share. You pay for that bucket's storage and requests, and nothing is cleaned up for you, so add a lifecycle rule to expire old cache objects.
  • Both: pass --cache and --cache-xz together.
  • Metadata TTL: --metadata-ttl 300 keeps file details for five minutes (or indefinite for data that stays fixed). With caching on, it defaults to one minute.

One security point: the local cache stores object contents unencrypted on disk, so lock down the cache folder's permissions, and give the shared cache bucket to Mountpoint clients only.

Jake: "The dashboard rereads the same twelve files every minute."

Ethan: "Local cache with a five-minute TTL. That cuts your GET requests to almost nothing."

Mountpoint performance tuning

Mountpoint is fast out of the box, but a few settings decide whether you get your instance's full bandwidth:

  • --maximum-throughput-gbps: the network throughput Mountpoint aims for. On EC2 it defaults to the instance's available bandwidth; elsewhere it defaults to 10 Gbps. The limit is shared by everything one Mountpoint process does, so if you run several mounts on one machine, lower each one's value so they do not compete.
  • --max-threads: how many file or directory operations can run at once, 16 by default. Raise it for workloads with many small files read in parallel.
  • --read-part-size and --write-part-size: the size of each S3 request, 8 MiB by default. Larger parts suit very large sequential files; the default is right for most work.
  • --memory-target: a target in MiB for Mountpoint's total memory use, minimum 512. Lower it on small instances where Mountpoint competes with your application; it also limits how many files can be open for writing at once.

The bigger factors are outside Mountpoint. Instance network bandwidth sets the ceiling, so a small instance cannot reach high throughput whatever the flags. Parallelism matters more than any single setting: many processes reading many files beat one process reading one file. And keep the instance in the same Region as the bucket, where transfer is fastest and free. To measure, time a large copy with dd or run fio with several parallel jobs reading from the mount, before and after a change.

Mountpoint logs and debugging

When a mount misbehaves, the logs say why. Mountpoint writes INFO and more serious messages to syslog by default (versions up to 1.20 logged only warnings and errors). On systemd distributions, read them with:

journalctl -e SYSLOG_IDENTIFIER=mount-s3

# or write logs to files, with more detail, while you debug
mount-s3 shop-reports-bucket ~/reports -l /var/log/mountpoint --debug
  • -l / --log-directory: write logs to files instead of syslog; the best choice with --debug.
  • --debug: verbose Mountpoint logs, showing each operation and S3 request.
  • --debug-crt: very verbose logs from the underlying AWS Common Runtime; use only with a log directory and only while chasing a network or credentials problem.
  • --log-metrics: periodic metrics such as request counts and throughput, useful when tuning.

Log files are not rotated or cleaned up automatically, so add a logrotate rule if you leave file logging on, or switch back to syslog once the problem is solved.

Mounting a bucket that belongs to another AWS account

Mounting a bucket owned by another account works like any cross-account S3 access, with two sides to configure:

  1. In the bucket owner's account: a bucket policy that allows your role s3:ListBucket on the bucket and the object actions you need on its objects; if the bucket uses a customer managed KMS key, the key policy must allow your role to decrypt (and generate data keys for writing).
  2. In your account: the instance role's IAM policy must allow the same actions on that bucket's ARNs.
  3. On the mount: add --expected-bucket-owner 111122223333 with the owner's account ID. If the bucket is not owned by that account, the mount fails with Access Denied instead of silently reading from or writing to someone else's bucket with the same name.

Objects written into another account's bucket are owned by the bucket owner when the bucket uses the recommended "bucket owner enforced" setting, which avoids the old problem of the owner being unable to read your uploads. Our cross-account access guide walks through the policies.

Mountpoint with S3 Express One Zone directory buckets

Directory buckets in the S3 Express One Zone storage class (named like reports--usw2-az1--x-s3) live in a single Availability Zone and answer with single-digit millisecond latency. With Mountpoint they unlock features that normal buckets lack: renaming files within the bucket, and appending to existing files with --incremental-upload. That makes them a good fit for scratch data, intermediate files in a pipeline, and logs written as they happen.

The trade-offs: data lives in one Availability Zone, so put your instances in the same zone for the latency benefit and keep anything you cannot afford to lose in a regular bucket as well; prices are higher per GB stored, though request prices are lower. Directory buckets also serve as Mountpoint's shared cache with --cache-xz; Mountpoint never deletes what it caches there, so set a lifecycle rule to expire old cache objects.

Mountpoint in Docker and on ECS

Mountpoint uses FUSE, which needs access to the /dev/fuse device and extra privileges that containers do not get by default. There are three ways to give containers access to a bucket, in order of preference:

  • On Kubernetes: the Mountpoint CSI driver (above), which runs Mountpoint for you and presents volumes to pods.
  • Mount on the host, bind-mount into the container: run Mountpoint on the EC2 host (with fstab at boot), then mount that folder into containers as a volume. On ECS with the EC2 launch type, this is a host volume in the task definition. The containers need no special privileges.
  • Run Mountpoint inside the container: possible, but the container needs the FUSE device and the SYS_ADMIN capability (for Docker, typically --device /dev/fuse --cap-add SYS_ADMIN, and on some hosts a relaxed security profile). That widens what the container can do, so prefer the host-mount pattern where you can. Mountpoint can also accept an already-open FUSE file descriptor, which allows unprivileged mounts in container setups built for it.

On Fargate none of these work, since Fargate does not expose FUSE; use an S3 Files volume there instead.

When not to mount S3 at all

A mount is the right tool when a program insists on a folder path and cannot be changed. When you control the code, other routes are often simpler, cheaper and easier to debug:

  • Your own code: read and write objects with the SDK (boto3 in Python, the AWS SDK in other languages). You get exact control over requests, retries and errors, with no FUSE layer in between.
  • Scheduled copies: aws s3 sync on a timer is hard to beat for "keep this folder in step with the bucket", especially when the folder is read many times between changes.
  • Big data tools: Spark, Athena, pandas and DuckDB read s3:// paths directly and parallelize far better than going through a mounted folder.
  • Uploads from users: presigned URLs let browsers upload straight to S3 without your servers handling the file at all.

Mount when a third-party application, a legacy script or a tool such as a media encoder needs a path, or when many readers need the same large files without each keeping a copy. In those cases Mountpoint (or S3 Files for full file behavior) saves real work.

Mounting S3 on servers outside AWS

Mountpoint runs on any supported Linux machine, not just EC2: an office server, a lab workstation or a machine in another cloud. It talks to S3's normal HTTPS endpoints, so no VPN or Direct Connect is required, although they help with speed and keep traffic private. Two differences from EC2 matter:

  • Credentials: there is no instance profile, so use IAM Roles Anywhere (short-term credentials from your own certificates) where you can, or an IAM user's access keys stored with restrictive file permissions as a last resort. Remember that fstab mounts run as root and read root's credentials.
  • Cost and speed: reading from S3 to a machine outside AWS is billed as data transfer out, and your internet link sets the ceiling. Turn on the local cache for anything read more than once, and mount read-only where possible so nothing writes back over a slow line.

For an office that wants a shared drive rather than a Linux mount, an S3 File Gateway VM is usually the better fit, since it caches locally and speaks SMB to desktops.

Upgrading or removing Mountpoint

Mountpoint releases often, and newer versions bring speed and compatibility fixes. On Amazon Linux 2023, sudo dnf upgrade mount-s3 fetches the latest package; elsewhere, download the newest RPM or DEB from the same URLs as the install and install it over the old one. A running mount keeps using the old binary until you unmount and mount again, so plan the upgrade with a remount (or a reboot for fstab mounts) and check mount-s3 --version afterwards. To remove it, unmount every bucket, delete its fstab lines, and uninstall the package with dnf remove mount-s3 or apt-get remove mount-s3.

On fleets, bake the Mountpoint version into your machine image or pin it in your configuration tool, and roll upgrades out to a few instances first. Read the release notes for changed defaults (such as the logging level change in version 1.21) so a new version does not surprise your monitoring.

Jake: "We upgraded and nothing changed."

Ethan: "The mount was still running the old binary. Unmount, mount again, then check the version."

Amazon S3 Files: a full file system on top of your bucket

Mountpoint's limits (no edits, no renames, no locks) are fine for data pipelines and painful for ordinary applications. Amazon S3 Files, generally available since April 7, 2026, fills that gap: it creates a shared NFS file system linked to a new or existing bucket, built on Amazon EFS technology. Your apps get normal file behavior with close-to-open consistency, changes sync back to the bucket automatically, and other tools keep using the same data as S3 objects at the same time.

The setup is different from Mountpoint, because there is a managed service in the middle:

  1. Create the S3 file system for your bucket (optionally scoped to a prefix) in the S3 console or CLI. The bucket must be a general purpose bucket with versioning turned on and SSE-S3 or SSE-KMS encryption.
  2. Create a mount target in each Availability Zone you will mount from, with a security group that allows NFS on port 2049 from your instances.
  3. Give the instance a role through an instance profile with the S3 Files permissions.
  4. Install the client: the amazon-efs-utils package, which adds an s3files mount type.
  5. Mount it:
    sudo mkdir /mnt/s3files
    sudo mount -t s3files fs-0123456789abcdef0:/ /mnt/s3files
    df -h /mnt/s3files
  6. Mount at boot with fs-0123456789abcdef0:/ /mnt/s3files s3files _netdev,nofail 0 0 in /etc/fstab.

The connection always uses TLS and IAM authentication. Supported clients are Linux only (Amazon Linux 2 and 2023, RHEL 8 and 9, Ubuntu 20.04 to 24.04, SLES 15, openSUSE Leap), and the compute must be in the same VPC (or reach it through peering or a transit gateway). Unlike Mountpoint, it also works from Lambda functions, from ECS tasks on Fargate or ECS Managed Instances (not the EC2 launch type), and from EKS through the EFS CSI driver.

Costs work differently too. Recently used files are loaded onto fast file system storage, and you pay for that active portion plus file system access; files you have not touched for a set period (30 days by default, 1 to 365) drop off the fast storage and stay in S3. Large reads of 1 MiB or more stream straight from S3 at normal GET request prices. For a bucket where only a small slice is busy, that is far cheaper than copying everything into a separate file system.

Mount S3 in Kubernetes with the Mountpoint CSI driver

On Amazon EKS, the Mountpoint for Amazon S3 CSI driver (EKS add-on name aws-mountpoint-s3-csi-driver) lets pods use a bucket as a volume. Things to know first:

  • Static provisioning only: you point a PersistentVolume at an existing bucket; it cannot create buckets.
  • No Fargate, no Windows containers and no EKS Hybrid Nodes. Pods on EC2 worker nodes are fine.
  • Version 2 (August 2025) runs Mountpoint in its own pods, lets pods on a node share one mount, supports EKS Pod Identity as well as IRSA, and adds SELinux support and caching shared between pods.
  • The same file system limits as Mountpoint apply inside the pod.
apiVersion: v1
kind: PersistentVolume
metadata:
  name: reports-pv
spec:
  capacity:
    storage: 1200Gi        # required by Kubernetes, ignored by the driver
  accessModes:
    - ReadWriteMany
  storageClassName: ""     # empty for static provisioning
  claimRef:
    namespace: default
    name: reports-claim
  mountOptions:
    - allow-delete
    - region eu-west-2
    - prefix daily/
  csi:
    driver: s3.csi.aws.com
    volumeHandle: reports-volume
    volumeAttributes:
      bucketName: shop-reports-bucket
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: reports-claim
spec:
  accessModes:
    - ReadWriteMany
  storageClassName: ""
  resources:
    requests:
      storage: 1200Gi
  volumeName: reports-pv

If the claim stays Pending, the usual causes are the same as for other CSI drivers: the add-on is not installed, or its role lacks permissions. Our EBS CSI PVC pending guide walks through that pattern. S3 Files also mounts in EKS, through the EFS CSI driver, when pods need full file behavior; on Fargate, S3 Files volumes on ECS are the supported route.

Mount an S3 bucket as a drive on Windows

Windows File Explorer cannot map an S3 bucket by itself, and Mountpoint is Linux only. Two good routes:

1. rclone mount with WinFsp (free, a drive letter on your PC or server):

  1. Install WinFsp (the Windows FUSE layer) from its official site, then install rclone, for example with winget install Rclone.Rclone.
  2. Create a remote that uses your normal AWS credentials (environment variables, the shared credentials file, or an instance role on EC2):
    rclone config create s3aws s3 provider AWS env_auth true region eu-west-2
  3. Mount the bucket on a drive letter:
    rclone mount s3aws:shop-reports-bucket S: --vfs-cache-mode full --network-mode
  4. Keep the window open: the mount lasts as long as rclone runs. To mount at sign-in or boot, start the same command from Task Scheduler or run it as a service.

To keep the drive mounted without a visible window, add --no-console and start the command from Task Scheduler with the trigger "At log on" (for your user) or "At startup" (for a server, running as an account that has the rclone config). Two more options help on desktops: --vfs-cache-max-size 20G stops the local cache from filling your disk, and --dir-cache-time 5m controls how long folder listings are reused before rclone asks S3 again, which trades freshness for fewer LIST requests.

--vfs-cache-mode full matters: without a cache mode, programs like Excel and Word fail because they want to edit files in place. With it, rclone works on a local copy and uploads the result. --network-mode makes it show as a network drive in This PC. Empty folders tend to disappear, because S3 has no real folders.

2. S3 File Gateway (for a team): AWS Storage Gateway runs as a virtual machine on VMware, Hyper-V or KVM in your office, or on EC2, and offers an SMB or NFS share whose files are stored one-to-one as objects in your bucket, with a local cache for speed. Users map it like any other share, with Active Directory permissions. You pay for the gateway's writes to AWS (capped at $125 per gateway a month, with the first 100 GB per account free) plus normal S3 storage and requests, and you run the VM. For a Windows file server that does not need to be S3 at all, FSx for Windows File Server is the managed choice.

A few paid desktop apps also map S3 as a drive with a friendlier setup; they suit one person's PC more than a server.

Mount S3 on a Mac

Mountpoint does not support macOS, and S3 Files is Linux only. On a Mac, rclone is the practical option. rclone mount needs a FUSE layer (macFUSE, which needs a system extension, or FUSE-T, which does not), and rclone nfsmount uses the Mac's built-in NFS client instead, so you install nothing extra. Use the same remote and --vfs-cache-mode full as on Windows, with a folder such as ~/s3 as the mount point:

brew install rclone
rclone config create s3aws s3 provider AWS env_auth true region eu-west-2
mkdir ~/s3
rclone nfsmount s3aws:shop-reports-bucket ~/s3 --vfs-cache-mode full

For developers on a Mac who only need to work with files in a bucket from scripts, the AWS CLI's aws s3 sync or a desktop S3 browser may be simpler than a mount; a mount is worth it when an app expects a normal folder path. Credentials come from the same ~/.aws files the CLI uses when the remote is set with env_auth true.

Finder then shows the bucket like any other folder. Unmount with umount ~/s3 when you are done, and expect large folders to open slowly the first time, because each one is an S3 listing.

s3fs-fuse vs Mountpoint vs rclone vs S3 Files

s3fs-fuse is the older open-source option, still in most package managers (sudo apt install s3fs, or from EPEL on RHEL). It fakes more file system features: it can rename by copying and deleting, store chmod and owner values in object metadata, and edit files by downloading and re-uploading the whole object. That is convenient and slow for large files. Mount it with an instance role like this:

s3fs shop-reports-bucket /data/reports -o iam_role=auto -o allow_other

# /etc/fstab
shop-reports-bucket /data/reports fuse.s3fs _netdev,allow_other,iam_role=auto 0 0
 MountpointS3 Filess3fs-fuserclone mount
Made byAWS, open sourceAWS, managed serviceCommunityCommunity
Runs onLinuxLinux EC2 in your VPC, Lambda, ECS on Fargate, EKSLinux, macOSWindows, macOS, Linux
Read speedVery highHigh; big reads stream from S3ModerateGood with cache
Edit in place, renameNo (rename on directory buckets only)YesYes, by re-uploadingYes, with the VFS cache
Shared writersNo coordinationYes, NFS semanticsNo coordinationNo coordination
Extra costNone; S3 requestsActive data and accessNone; S3 requestsNone; S3 requests
Best forAnalytics, ML training, logsOrdinary apps on S3 dataOlder scripts, small filesDesktops, Windows

Two writers on Mountpoint, s3fs or rclone can overwrite each other with no warning; S3 keeps whichever upload finishes last. If several machines write the same files, S3 Files or EFS is the right tool.

How much mounting S3 costs

Mountpoint, s3fs and rclone are free software, so the bill is ordinary S3: storage, plus a charge for every request. A mount hides those requests, which is how a harmless-looking ls -R or a backup tool scanning the folder runs up LIST and GET charges. Rough guide:

  • Listing a folder costs LIST requests (one per 1,000 keys); tools that list repeatedly multiply that.
  • Reading a file costs GET requests; large files are read in several ranged GETs.
  • Writing a large file is a multipart upload, one PUT per part.
  • From EC2 in the same Region, data transfer to and from S3 is free. From your office or another Region, downloads are charged as data transfer out.
  • Caching cuts repeated GETs; a shared Express One Zone cache adds its own storage and request charges.

For the arithmetic and the storage classes that change it, see our S3 request pricing guide and storage classes guide. S3 Files adds the fast-storage and access charges described above, and S3 File Gateway adds its write fee.

Jake: "Our backup agent crawls every mounted folder each night."

Ethan: "Exclude the mount. Otherwise you are paying to list a bucket you already have in S3."

Common mount errors and fixes

  • mount-s3: command not found: the package is not installed, or for the tarball install, /opt/aws/mountpoint-s3/bin is not on your PATH.
  • Access Denied or 403 when mounting: the role lacks s3:ListBucket on the bucket ARN (not the /* ARN), or a bucket policy, VPC endpoint policy or KMS key blocks it. Our GetObject AccessDenied guide walks the layers.
  • No credentials found: no instance profile, keys or role profile visible to the user running the mount (root, for fstab), or you are using an SSO profile.
  • Permission denied for another user, or for root: mount with --allow-other and add user_allow_other to /etc/fuse.conf.
  • Operation not permitted when saving an existing file: overwrites are off (add --allow-overwrite), or the program edits in place or renames, which Mountpoint does not do.
  • rm fails: deletes are off; add --allow-delete.
  • fuse: device not found or /dev/fuse missing: FUSE is not loaded, or you are inside a container that does not expose it (Fargate cannot).
  • Transport endpoint is not connected: the mount process died. Run sudo umount -l /data/reports, mount again, and check the logs.
  • fstab mount does nothing at boot: one of _netdev, nosuid, nodev is missing, or root has no credentials; systemctl status data-reports.mount shows the reason.
  • New objects missing from the listing: a long --metadata-ttl is serving the cached listing; shorten it, or open the file directly.
  • Reading some files fails: they are in Glacier Flexible Retrieval or Deep Archive; restore them first.

Mountpoint writes its logs to syslog by default, so journalctl -t mount-s3 (or your syslog file) shows them; run with -f --debug in a terminal to watch requests live while you fix things.

Frequently asked questions

How do I mount an S3 bucket on Linux?

Install Mountpoint for Amazon S3 and run mount-s3 bucket-name /path/to/folder; for edits and renames, use Amazon S3 Files instead.

Can I mount an S3 bucket as a drive on Windows?

Not with Mountpoint, which is Linux only. Use rclone mount with WinFsp for a drive letter, or an S3 File Gateway SMB share.

Is Mountpoint for Amazon S3 free?

Yes. It is open-source software; you pay only the normal S3 storage, request and data transfer charges.

Does Mountpoint for Amazon S3 work on Fargate?

No. Fargate cannot run FUSE mounts and the CSI driver does not support Fargate; use an Amazon S3 Files volume for ECS tasks on Fargate.

How do I mount S3 at boot with fstab?

Add a line like s3://bucket/ /mnt/bucket mount-s3 _netdev,nosuid,nodev,nofail,rw 0 0 to /etc/fstab, then run systemctl daemon-reload and mount -a.

Can I edit files on a Mountpoint S3 mount?

You can create new files and replace whole files with --allow-overwrite, but not edit part of a file or rename files on general purpose buckets.

What is Amazon S3 Files?

A managed NFS file system linked to an S3 bucket, generally available since April 2026, that gives full file operations and syncs changes back to the bucket.

What permissions does Mountpoint need?

s3:ListBucket on the bucket and s3:GetObject on objects for reading; add s3:PutObject, s3:AbortMultipartUpload and s3:DeleteObject for writing.

How do I mount S3 in EKS?

Install the aws-mountpoint-s3-csi-driver add-on and create a static PersistentVolume with driver s3.csi.aws.com and your bucketName.

Can Mountpoint for Amazon S3 run on macOS?

No. On a Mac, use rclone mount with macFUSE or FUSE-T, or rclone nfsmount.

What is the difference between s3fs and Mountpoint?

Mountpoint is faster and backed by AWS but skips edits and renames; s3fs emulates more file features by re-uploading whole objects, which is slower.

Why do other users get permission denied on the mount?

Only the mounting user has access by default; add user_allow_other to /etc/fuse.conf and mount with --allow-other.

How do I mount only one folder of a bucket?

Use --prefix folder/ with mount-s3, or s3://bucket/folder/ in fstab; the prefix must end with a slash.

Does mounting S3 cost extra?

The tools are free, but every list, read and write is an S3 request, so scanning a mounted folder adds request charges.

Can Mountpoint read Glacier objects?

Not from Glacier Flexible Retrieval, Glacier Deep Archive or the Intelligent-Tiering archive tiers until they are restored.

How do I unmount an S3 bucket?

Run umount with the folder path; if the process died and shows Transport endpoint is not connected, use umount -l.

Can I use Mountpoint on a server outside AWS?

Yes, on any supported Linux; use IAM Roles Anywhere or access keys for credentials, and expect data transfer out charges for reads.

How do I upgrade Mountpoint for Amazon S3?

Install the newest package (dnf upgrade mount-s3 on Amazon Linux 2023), then unmount and mount again so the new version is used.

Should I mount S3 or use the SDK?

Use the SDK or aws s3 sync when you control the code; mount when an application needs a normal folder path and cannot be changed.

How do I make Mountpoint faster?

Use an instance with more network bandwidth, read many files in parallel, raise --max-threads for many small files, and add a local cache for data you reread.

Where are Mountpoint logs?

In syslog by default; on systemd run journalctl -e SYSLOG_IDENTIFIER=mount-s3, or use -l with a folder and --debug for detailed file logs.

Can I mount an S3 bucket from another AWS account?

Yes, with a bucket policy (and KMS key policy if needed) for your role, matching IAM permissions, and --expected-bucket-owner set to the owner's account ID.

Can I use Mountpoint inside a Docker container?

It needs /dev/fuse and the SYS_ADMIN capability; mounting on the host and bind-mounting the folder into the container is the safer pattern.

Does Mountpoint support renaming files?

Only on S3 Express One Zone directory buckets, which also support appends with --incremental-upload; general purpose buckets do not.

How do I keep an rclone S3 drive mounted on Windows?

Run rclone mount with --no-console from Task Scheduler at log on or at startup, with --vfs-cache-mode full.

Is the Mountpoint cache encrypted?

No. The local cache stores object data unencrypted, so restrict access to the cache folder.

For readers preparing for an AWS certification

We have been covering AWS certifications alongside our fixes, so here is a short extra for readers studying for one. If you only came for the fix, you already have it; feel free to skip this part.

Exams that test this topic: Solutions Architect – Associate (SAA-C03) • CloudOps Engineer – Associate (SOA-C03) • Developer – Associate (DVA-C03 from December 1, 2026).

Here are the exam-style questions this topic tends to raise, with the reasoning you need to answer them:

1. Which open-source AWS client mounts an S3 bucket as a local file system on Linux for high-throughput reads?

Mountpoint for Amazon S3.

2. An application on several instances must edit files in place, rename them and share writes on data kept in S3. What fits?

A shared file system linked to the bucket, such as Amazon S3 Files, rather than a FUSE client.

3. Office users need an SMB share whose files are stored as objects in S3. Which service provides it?

AWS Storage Gateway, S3 File Gateway.

4. How should an EC2 instance get credentials to mount a bucket?

An IAM role attached through an instance profile.

5. Which permission lets a client list the objects in a bucket, and which resource ARN does it apply to?

s3:ListBucket, on the bucket ARN (not the /* object ARN).

6. What protects data in a bucket against accidental deletes made through a mount?

S3 Versioning, so deleted objects can be restored.

7. Is data transfer between S3 and an EC2 instance in the same Region charged?

No; same-Region transfer between S3 and EC2 is free.

Exam codes and dates change often. Our AWS certification changes page keeps track of what is current.

πŸ“š ALSO READ

Where to go next, from EFS and FSx to S3 costs and access errors:

📌 Bookmark this; the which-tool table and the fstab line are the parts you will come back for.

The hourly sync job is gone. Jake's server mounts the reports prefix read-only at boot, the script reads the CSVs exactly as before, and the summary file goes to a second, writable mount with --allow-overwrite. The disk has stayed at a third full for a month, and the dashboard's repeated reads come from a small local cache instead of S3.

📌 If you keep one line from this page

On Linux, mount S3 with Mountpoint (mount-s3 bucket /folder) for fast reads and new files, use Amazon S3 Files when apps need edits, renames or shared writes, and use rclone with WinFsp or an S3 File Gateway on Windows.

_netdev,nosuid,nodev in fstab · --allow-other for other users · versioning before --allow-delete.

Revision note. Written October 20, 2026. A mounted bucket is still a bucket: every file operation is an S3 request.

Related