Re: [BUG] ext4/fscrypt: memory corruption after enabling "encrypt" on a mounted fs (large folios + bounce page), 7.0

Re: [BUG] ext4/fscrypt: memory corruption after enabling "encrypt" on a mounted fs (large folios + bounce page), 7.0

[Eric Biggers]

On Sun, Oct 25, 2026 at 04:41:16PM -0700, Joshua Abata wrote:
> Hi,
> 
> Enabling the ext4 "encrypt" feature with tune2fs on a *mounted* filesystem
> and then writing through an fscrypt v2
> policy corrupts kernel memory on 7.0 (Ubuntu 7.0.0-38-generic). It
> reproduces in seconds on an untainted kernel in a
> plain QEMU VM; the same steps with an unmount/remount after tune2fs are
> fine.
> 
> Reproducer (as root, any 7.0 x86_64 with ext4 large folios):

7.0 is no longer a supported kernel version upstream.  It was not chosen
as an LTS kernel, and the last 7.0.y stable release occurred on June 27
with Linux v7.0.14.

This sounds like the bug that was already fixed in 7.2 by
"ext4: don't enable large folios on encrypted inodes"
(https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=linux-7.2.y&id=40554d1221e3d2ff2e0d192b74a05d89eb3620e7) and in 7.3 by other changes.

If Ubuntu is using an unsupported kernel version that is missing this
fix, then you would need to report that to Ubuntu.

- Eric