Compelling reasons to switch ? [LWN.net]

Compelling reasons to switch ?

Posted Jan 23, 2026 17:15 UTC (Fri) by Lionel_Debroux (subscriber, #30014)
Parent article: Filesystem medley: EROFS, NTFS, and XFS

On my computer, the very first public release of ntfs-3g ran out of steam for files larger than several hundred MBs, but that was fixed quickly. Since then, which was nearly two decades ago, ntfs-3g has been reliable IME.

Honest questions:
* for most users, with the ever increasing focus on security, what are the highly compelling reasons to use kernel-mode drivers like ntfs3, or that new version of ntfs, instead of trusty old FUSE ntfs-3g ? Performance, really ?
* for security-minded users aiming at reducing the attack surface without resorting to building and signing their own kernels, why should their hardened deployments of standard Linux distros not add into some /etc/modprobe.d/* file:
install ntfs /bin/true
right next to
install ksmbd /bin/true
and a number of similar lines for disabling rarely used filesystems and/or repeat security offenders ?

to post comments

Compelling reasons to switch ?

Posted Jan 23, 2026 17:59 UTC (Fri) by pizza (subscriber, #46) [Link] (2 responses)

> Performance, really ?

"Security" is not the focus of most users... as opposed to things like "don't eat my data".. and yes, "performance".

Filesystems are routinely benchmarked on "performance". But you can't sell "security" the same way.

Compelling reasons to switch ?

Posted Jan 24, 2026 8:08 UTC (Sat) by Lionel_Debroux (subscriber, #30014) [Link]

Nearly 20 years ago, after the first few releases of ntfs-3g fixed the initial performance issues with large files, ntfs-3g was much faster than HDDs anyway. When dealing with a PCIe Gen5 NVME SSD sporting millions of IOPS and a transfer rate in the GB/s range, there could be a speed and/or CPU consumption difference between a kernel space filesystem and a user space filesystem, right. However, it remains to be seen whether the latter isn't good enough for _most users_ on those metrics.

I guess I'll wait for benchmarks to be posted on e.g. Phoronix.

Compelling reasons to switch ?

Posted Jan 29, 2026 15:25 UTC (Thu) by LionsPhil (subscriber, #121073) [Link]

I don't think that really applies when it comes to file system implementations. For the vast majority of users you're not presented with a choice; you're going to get whatever your GUI filer autoselects when you click the icon for the drive, or mount autoselects when you just do "mount /dev/sdb1 /tmp/whatever".

Which is putting the decision in the hands of distro, kernel, util-linux, KIO, whatever-GNOME's-is-called, etc. maintainers. I assume that's going to be more influenced by "which one gets me the fewest severe bug reports?"

Add to that that a lot of filesystem performance pressure comes from the server role, but...hopefully the number of people choosing to use NTFS on Linux servers is small. :D As a reverse-engineered-for-interoperability filesystem, this is more the remit of client systems supporting being able to read that Windows-world external drive you plugged in. And, consequently, at *more* risk of vulnerabilities in the filesystem driver.

(I just wish ntfs-3g supported sparse files. NTFS itself does.)

Compelling reasons to switch ?

Posted Jan 23, 2026 18:23 UTC (Fri) by marcel.oliver (subscriber, #5441) [Link] (7 responses)

I experienced a very "interesting" bug that might well give a compelling reason for a more robust NTFS implementation: A graduate student of mine was doing, fairly recently, a parameter study of some numerical modeling problem. He was controlling the actual simulation (some Fortran code) from a Python script that would first generate a working directory with a deep path, each of the many parameters encoding one level of subdirectories with the actual parameter value going into the directory name. He was running all of this on a mildly beefy departmental compute server, dumping the output onto a big USB hard drive which, no surprise, came formatted with NTFS. For the analysis, he plugged in the hard drive into his laptop and used different Python scripts to read a single data point from each of the simulations for his study. Both server and laptop had some recent Ubuntu, so presumably used ntfs-3g.

Now it turned out that his graphs had strange outliers. On closer inspection, the "outliers" were actually at different locations for different runs of the script, while the data on disk was completely unchanged. Excluding all other rational explanations and given that the nesting of directories was pretty deep and I was not sure that all characters used where actually legal and unambiguous on NTFS, I concluded that this might be a filesystem issue, with the filesystem randomly and quietly mixing up paths. So I advised to simply reformat the drive in ext4 and recompute everything. Immediately, the "outliers" were gone.

I still think we were extremely lucky that ntfs-3g's path resolution was wrong in a non-deterministic fashion, otherwise the problem space would have been much larger and we would likely not have suspected the filesystem as the culprit. But from this experience, I am very reluctant to touch NTFS from Linux in any way.

Compelling reasons to switch ?

Posted Jan 23, 2026 20:26 UTC (Fri) by rgmoore (✭ supporter ✭, #75) [Link] (6 responses)

I experienced a very "interesting" bug that might well give a compelling reason for a more robust NTFS implementation ... But from this experience, I am very reluctant to touch NTFS from Linux in any way.

I think the second point is the right one. Having an NTFS driver is cool because it helps with interoperability with Windows systems, but that should be its exclusive use. If every system that's going to use your disk runs Linux, they should use a Linux-native filesystem. This isn't intended as a dig against NTFS, just a comment on the degree of testing the different filesystem code is likely to have.

Compelling reasons to switch ?

Posted Jan 25, 2026 18:21 UTC (Sun) by eru (subscriber, #2753) [Link] (5 responses)

Sadly, interoperability with Windows is a common use-case, and it is bidirectional. I like to archive stuff offline on removable USB HD:s, and wish them to be both readable and writable from both Linux and Windows systems. NTFS is really the only option, because Windows has no up-to-date and well-maintained implementations for any Linux file systems, and FAT just does not cut it any more with today's large files. So all work on better NTFS support in Linux is welcome.

Compelling reasons to switch ?

Posted Jan 25, 2026 19:12 UTC (Sun) by jem (subscriber, #24231) [Link] (4 responses)

ExFAT and UDF are two filesystems that are suitable for removable drives, and can be used to transfer large files between Linux and Windows (and macOS too.)

Compelling reasons to switch ?

Posted Jan 25, 2026 20:10 UTC (Sun) by marcH (subscriber, #57642) [Link] (2 responses)

UDF looks nice in but I'm afraid it's poorly maintained - including on Windows!

exFAT is probably a better choice now.

Compelling reasons to switch ?

Posted Jan 26, 2026 2:08 UTC (Mon) by willy (subscriber, #9762) [Link] (1 responses)

Jan Kara maintains UDF, and I'm not aware of any outstanding problems or patches that are unduly delayed. Sure, there's not a huge flux of patches, but that's to be expected, I think.

It'd be nice if it were converted to use iomap instead of buffer_heads. If anyone's looking for a filesystem project ...

Compelling reasons to switch ?

Posted Jan 26, 2026 9:10 UTC (Mon) by chris_se (subscriber, #99706) [Link]

> Jan Kara maintains UDF, and I'm not aware of any outstanding problems or patches that are unduly delayed.

The main reason for not using UDF as a data exchange format is not the Linux support in my eyes, but the rather non-optimal support on other platforms. Reading from UDF works on other platforms from my experience, but in the past I've had trouble writing to UDF filesystems on Windows - sometimes it just didn't work, with no apparent reason for the failure.

UDF is nice in theory, but in practice exFAT is probably what's here to stay for simple data exchange.

Compelling reasons to switch ?

Posted Jan 25, 2026 20:41 UTC (Sun) by Cyberax (✭ supporter ✭, #52523) [Link]

ExFAT is not journaled, making it less suitable for something like an internal HDD/SSD that is used to store data on a dual-boot system.

Compelling reasons to switch ?

Posted Feb 24, 2026 5:29 UTC (Tue) by quotemstr (subscriber, #45331) [Link]

Shouldn't file data sharing be a VFS facility, not something specific to EROFS?