Disk trimming is the procedure by which the operating system informs the underlying storage device of which storage blocks are no longer in use. It does this by issuing an
ATA_TRIM command for the block. This is also known as a
discard. In this way, the storage device can perform garbage collection of the unused blocks and internally prepare them for reuse. SSDs in general benefit from this, while HDDs do not.
In a Linux system running on bare metal, this is relatively straight-forward. When instructed by the operating system, discards are issued by the file-system driver directly to the storage driver and then to the SSD.
In Qubes, this gets more complex due to virtualization, LUKS, and LVM (and thin pools on R4.0 and up). If you run
fstrim --all inside a TemplateVM, in a worst case the
discard can follow a path like:
OS -> File-system Driver -> Virtual Storage Driver -> Backend Storage Driver -> LVM Storage Driver -> LUKS Driver -> Physical Storage Driver -> Physical Storage Device
If discards are not supported at any one of those layers, it will not make it to the underlying physical device.
There are some security implications to permitting TRIM (read for example this article), but in most cases not exploitable. Conversely, TRIM can improve security against local forensics when using SSDs, because with TRIM enabled deleting data (usually) results in the actual data being erased quickly, rather than remaining in unallocated space indefinitely. However deletion is not guaranteed, and can fail to happen without warning for a variety of reasons.
In all versions of Qubes, you may want to set up a periodic job in
dom0 to trim the disk. This can be done with either systemd (weekly only) or cron (daily or weekly). You should also add this service or cron job to each VM you want the trim function to work in.
From a terminal as a regular user:
systemctl enable fstrim.timer systemctl start fstrim.timer
This can be done from a terminal as root, by creating a
/etc/cron.weekly). Add the following contents:
#!/bin/bash /sbin/fstrim --all
And mark it as executable with
chmod 755 /etc/cron.daily/trim.
Note Although discards can be issued on every delete inside
dom0 by adding the
discard mount option to
/etc/fstab, this option can hurt performance so the above procedure is recommended instead. However, inside App and Template qubes, the
discard mount option is on by default to notify the LVM thin pool driver that the space is no longer needed and can be zeroed and re-used.
If you are using Qubes with LVM, you may also want to set
issue_discards = 1 in
/etc/lvm/lvm.conf. Setting this option will permit LVM to issue discards to the SSD when logical volumes are shrunk or deleted. In R4.x, LVM Logical volumes are frequently deleted (every time a disposable VM is shut down, for example) so you may want to set
issue_discards = 1 if using an SSD, but see the article linked in the first section of this page. However, this is relatively rare in R3.x.
If you have enabled LUKS in dom0, discards will not get passed down to the storage device in R4.0. Note that this procedure is no longer needed in R4.1, as discards are enabled by default.
To enable TRIM support in dom0 with LUKS you need to:
Get your LUKS device UUID:
Add entry to
/etc/crypttab(replace luks-<UUID> with the device name and the <UUID> with UUID alone):
luks-<UUID> UUID=<UUID> none discard
rd.luks.options=discardto kernel cmdline (follow either GRUB2 or EFI, not both):
Rebuild grub config (
grub2-mkconfig -o /boot/grub2/grub.cfg), then
Rebuild initrd (
Rebuild initrd (
dracut -f /boot/efi/EFI/qubes/initramfs-$(uname -r).img $(uname -r))
Reboot the system.
To verify if discards are enabled you may use
dmsetup table(confirm the line for your device mentions “discards”) or just run
fstrim -av(you should see a
/followed by the number of bytes trimmed).
By default TRIM is not enabled for swap in both R4.0 and R4.1. To enable it add the
discard flag to the options for the swap entry in
/etc/fstab. This may or may not actually improve performance. If you only want the security against local forensics benefit of TRIM, you can use the
discard=once option instead to only perform the TRIM operation once during at boot.
To verify that TRIM is enabled, check
dmesg for what flags were enabled when the swap space was activated. You should see something like the following:
Adding 32391164k swap on /dev/mapper/qubes_dom0-swap. Priority:-2 extents:1 across:32391164k SSDscFS
s indicates that the entire swap device will be trimmed at boot, and
c indicates that individual pages are trimmed after they are no longer being used.