mbox series

[SRU,M,0/2] disable shiftfs

Message ID 20231005094913.17839-1-andrea.righi@canonical.com
Headers show
Series disable shiftfs | expand

Message

Andrea Righi Oct. 5, 2023, 9:49 a.m. UTC
BugLink: https://bugs.launchpad.net/bugs/2038522

[Impact]

Now that all the filesystems that we officially support have the
idmapped mounts capability we can get rid of shiftfs.

The benefit of this change is that we don't have to maintain an
out-of-tree filesystem anymore and we can completely rely on upstream
features.

[Test case]

lxd was the main user of shiftfs to compensate the lack of idmapped
mounts capability of certain filesystems, such as zfs / ceph, but now in
mantic also these two filesystem received the support for idmapped
mounts (support for zfs was introduced in 2.2.0~rc3 and for ceph see
LP: #2032959).

The lxd team provided a positive feedback, testing the latest 6.5 Mantic
kernel across all the supported filesystems with shiftfs disabled.

[Fix]

Disable shiftfs in the kernel config and enable unsafe idmapped mounts
by default (default=on).

[Regression potential]

The support for idmapped mounts for the ceph filesystem is not applied
upstream yet, so we may experience regressions in systems that are using
this filesystem. Moreover disabling shiftfs may trigger failures in our
testing (testing shiftfs capabilities will obviously fail) or break any
other user-space application that is relying on shiftfs (however to our
knowledge lxd was the only "official" user or shiftfs; for this reason
we may also see potential regressions in lxd).

Comments

Tim Gardner Oct. 5, 2023, 12:35 p.m. UTC | #1
On 10/5/23 3:49 AM, Andrea Righi wrote:
> BugLink: https://bugs.launchpad.net/bugs/2038522
> 
> [Impact]
> 
> Now that all the filesystems that we officially support have the
> idmapped mounts capability we can get rid of shiftfs.
> 
> The benefit of this change is that we don't have to maintain an
> out-of-tree filesystem anymore and we can completely rely on upstream
> features.
> 
> [Test case]
> 
> lxd was the main user of shiftfs to compensate the lack of idmapped
> mounts capability of certain filesystems, such as zfs / ceph, but now in
> mantic also these two filesystem received the support for idmapped
> mounts (support for zfs was introduced in 2.2.0~rc3 and for ceph see
> LP: #2032959).
> 
> The lxd team provided a positive feedback, testing the latest 6.5 Mantic
> kernel across all the supported filesystems with shiftfs disabled.
> 
> [Fix]
> 
> Disable shiftfs in the kernel config and enable unsafe idmapped mounts
> by default (default=on).
> 
> [Regression potential]
> 
> The support for idmapped mounts for the ceph filesystem is not applied
> upstream yet, so we may experience regressions in systems that are using
> this filesystem. Moreover disabling shiftfs may trigger failures in our
> testing (testing shiftfs capabilities will obviously fail) or break any
> other user-space application that is relying on shiftfs (however to our
> knowledge lxd was the only "official" user or shiftfs; for this reason
> we may also see potential regressions in lxd).
> 
> 
Acked-by: Tim Gardner <tim.gardner@canonical.com>
Stefan Bader Oct. 11, 2023, 7:59 a.m. UTC | #2
On 05.10.23 11:49, Andrea Righi wrote:
> BugLink: https://bugs.launchpad.net/bugs/2038522
> 
> [Impact]
> 
> Now that all the filesystems that we officially support have the
> idmapped mounts capability we can get rid of shiftfs.
> 
> The benefit of this change is that we don't have to maintain an
> out-of-tree filesystem anymore and we can completely rely on upstream
> features.
> 
> [Test case]
> 
> lxd was the main user of shiftfs to compensate the lack of idmapped
> mounts capability of certain filesystems, such as zfs / ceph, but now in
> mantic also these two filesystem received the support for idmapped
> mounts (support for zfs was introduced in 2.2.0~rc3 and for ceph see
> LP: #2032959).
> 
> The lxd team provided a positive feedback, testing the latest 6.5 Mantic
> kernel across all the supported filesystems with shiftfs disabled.
> 
> [Fix]
> 
> Disable shiftfs in the kernel config and enable unsafe idmapped mounts
> by default (default=on).
> 
> [Regression potential]
> 
> The support for idmapped mounts for the ceph filesystem is not applied
> upstream yet, so we may experience regressions in systems that are using
> this filesystem. Moreover disabling shiftfs may trigger failures in our
> testing (testing shiftfs capabilities will obviously fail) or break any
> other user-space application that is relying on shiftfs (however to our
> knowledge lxd was the only "official" user or shiftfs; for this reason
> we may also see potential regressions in lxd).
> 
> 

Acked-by: Stefan Bader <stefan.bader@canonical.com>
Stefan Bader Oct. 25, 2023, 2:45 p.m. UTC | #3
On 05.10.23 11:49, Andrea Righi wrote:
> BugLink: https://bugs.launchpad.net/bugs/2038522
> 
> [Impact]
> 
> Now that all the filesystems that we officially support have the
> idmapped mounts capability we can get rid of shiftfs.
> 
> The benefit of this change is that we don't have to maintain an
> out-of-tree filesystem anymore and we can completely rely on upstream
> features.
> 
> [Test case]
> 
> lxd was the main user of shiftfs to compensate the lack of idmapped
> mounts capability of certain filesystems, such as zfs / ceph, but now in
> mantic also these two filesystem received the support for idmapped
> mounts (support for zfs was introduced in 2.2.0~rc3 and for ceph see
> LP: #2032959).
> 
> The lxd team provided a positive feedback, testing the latest 6.5 Mantic
> kernel across all the supported filesystems with shiftfs disabled.
> 
> [Fix]
> 
> Disable shiftfs in the kernel config and enable unsafe idmapped mounts
> by default (default=on).
> 
> [Regression potential]
> 
> The support for idmapped mounts for the ceph filesystem is not applied
> upstream yet, so we may experience regressions in systems that are using
> this filesystem. Moreover disabling shiftfs may trigger failures in our
> testing (testing shiftfs capabilities will obviously fail) or break any
> other user-space application that is relying on shiftfs (however to our
> knowledge lxd was the only "official" user or shiftfs; for this reason
> we may also see potential regressions in lxd).
> 
> 

Applied to mantic:linux/master-next. Thanks.

-Stefan