vmexit's can cause the AMX registers to "misbehave" which can break
ZFS, even though ZFS doesn't use AMX at all.
This causes crashes and processes hanging forever in uninterruptible
sleep (the infamous D state) on Intel Xeon 4th gen HW, possible other
HW too, but we only got reports on Sapphire Rapids models.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Tested-by: Aaron Lauterer <a.lauterer@proxmox.com>
(cherry picked from commit 9e8946d4b9)
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
As among possibly other things, it addresses a patch that got
backported to kernel 6.2.8 too.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
As reported having set 'mountpoint=legacy' on the root dataset (for a
default install this would be 'rpool/ROOT/pve-1') for mounting through
e.g. /etc/fstab breaks booting in initramfs
The cherry-picked patch is already applied to zfs-2.1.8-staging
upstream so will be included in the next zfs point release.
reproduced the issue on a VM with ZFS on root by:
* booting into a PVE 7.3 iso (to have zfs and an inactive
rpool/ROOT/pve-1)
* `zpool import -ocachefile=none -N -f rpool
* `zfs set mountpoint=legacy rpool/ROOT/pve-1`
* `mount -t zfs rpool/ROOT/pve-1 /mnt
* `echo "rpool/ROOT/pve-1 / zfs defaults 0 0" >> /mnt/etc/fstab`
* `zpool export rpool`
* reboot - the boot breaks with zfs-initramfs 2.1.7-pve1 but works
with this patch applied
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
While checking the current state of 2.1.6 we noticed that there were
some changes in debian-upstream [0] resulting from a bug-report in
zfs-upstream [1].
Our packages should be unaffected (they do not ship the
init-scripts in the first place).
Since the issue was fixed by zfs-upstream already on the
zfs-2.1.7-staging branch we should include it as well as it might save
users, who somehow got debian-upstream's zfs-packages installed on PVE
some trouble.
The other 2 changes seem small and isolated enough so including them
as well.
[0] https://salsa.debian.org/zfsonlinux-team/zfs/-/commits/master
[1] https://github.com/openzfs/zfs/issues/14010
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
the changes to zvol_open added to 2.1.2 (for coping with kernel
changes in 5.13) seem to have introduced a lock order inversion [0].
(noticed while reviewing the 2.0.6->2.0.7 changes (the patch was
applied after 2.1.2 was already tagged)
[0] https://github.com/openzfs/zfs/pull/12863
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
This commit updates Thomas' patch to deal with a 2.0 kernel module
with 2.1 arc_summary/arcstat
Tested by adding a cache-device to a zpool and running both commands
to verify no KeyError exception is thrown.
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
the patch fixes a potential panic on systems running ZFS > 2.0.0 and
is already queued for inclusion in 2.0.3 - see [0] for a related
github issue.
[0] https://github.com/openzfs/zfs/issues/11474
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
This patch addresses the problems some users experience when some zpools are
created/imported with cachefile (which then causes other pools not to get
imported during boot) - when our tooling creates a pool we explictly
instantiate the service with the pool's name, ensuring that it will get
imported by scanning.
Suggested-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
the last series of cherry-picks from salsa.d.o includes one patch for the
zfs-source, but the patch file does not contain ident-information.
This prevents it from being applied by import-patchqueue.
Fixed by adding ident based on the original patches commit message and
importing and exporting the patchqueue.
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Of the zedlet scripts shipped by upstream, a subset are enabled by
default, by creating symlinks in /etc/zfs/zed.d. These symlinks are
shipped in the zfs-zed package. dpkg, however, does not support
conffile handling of symlinks, and therefore any changes (removals) to
the symlinks are not preserved on package upgrade.
To address this policy violation, we:
1. During package build, create a list of enabled-by-default zedlets,
instead of creating symlinks.
2. On package removal, identify all enabled-by-default zedlets whose
symlinks do not exist (i.e., were removed by the user). This is done
by creating "whiteout" links to /dev/null in their place).
3. On package installation, create links to enabled-by-default zedlets
UNLESS there is already a file there (i.e., abort if there is a
whiteout link).
4. We also clean up broken symlinks to removed zedlets at package
postinst.
(cherry picked and adapted from 5cee380324d74e640d5dd7a360faba3994c8007f [0])
[0] https://salsa.debian.org/zfsonlinux-team/zfs.git
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
This reduces the required time for the configure step drastically,
from previous:
# time ./configure --with-config=kernel
-> 124.14s user 34.06s system 103% cpu 2:32.90 total
to now:
# time ./configure --with-config=kernel
-> 75.07s user 15.01s system 394% cpu 22.821 total
(152 seconds vs 22 seconds)
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
That's why it was guarded with the "HAVE_KERNEL_FPU_INITIALIZED"
defined in the first place..
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
