ran into an issue when clicking on the ESP of a VM while trying
single-file restore. Enabling the single option fixed the issue, and
got the partition successfully mounted.
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Link: https://lore.proxmox.com/20241126132602.576027-1-s.ivanov@proxmox.com
it's the more modern version, the legacy 'ntfs' one is already removed upstream
nowadays.. initial feedback on a test build was positive throughout - while it
still doesn't support all features of the Windows version, it does handle a lot
of backups better than the previous one..
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
* for the kernel-patch this includes a rename from 0003+0004 to
0001+0002
* for ZFS there was a change in upstream's autotools-setup - I
referenced the commit in the actual patch-file
minimally tested with a VM with a zfs-pool and an ext4 disk
* restore of a directory on ext4 containing 160MB of debian packages
as tar.zstd
* restore of a small folder (/root in a debian container) on zfs
both worked
restoring files from a Windows guest - worked, however there is an
independent issue with tpmstate not being found:
`given image 'drive-tpmstate0-backup.img.fidx' not found (400)`
directories with 10 million files also still cause the restore-shim to
run into OOM (but this is independent of the restore-image)
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Fixes a small regression in 0cec05b ("build initramfs: download
packages into per-flavor dirs"), which while taking a better
approach in general, missed that the debug package list is selected
with the assumption that it builds on top of the base list, but with
the split into two separate directories (to fix pre-caching), this
was lost.
While the packages selected pulled in a surprising amount of reverse
dependencies, it was still missing some like lvm2 tools (dmeventd and
co) and most notably busybox - which is really nice to have in the
debug image.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
whit dropping this in 0cec05b ("build initramfs: download packages
into per-flavor dirs"), the pre-caching was made effectively useless,
and building without internet during the build-step failed.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
this reduces the needed memory quite significantly (most extreme
example is libzpool from 17M to 4.3M, and init itself 15M->475K).
With this change the regular image is able to boot with
proxmox-file-restore and 192M.
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
commit 83c46462ac requires busybox to be
installed, which was only present in the debug image.
switching to busybox from busybox-static reduces the memory-footprint
quite a bit (699k vs. 1.9M) - and busybox only depends on libc6)
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
this should save a small bit of compiletime and should reduce the
space needed for the initramfs image
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
since libc6 in bookworm does not provide use '--enable-obsolete-rpc'
anymore and zfs needs it (building '--without-libtirpc' fails) - we
add it to both images.
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
commit f773602bd3 introduced pre-caching
packages in the source-package, but resulted in both initrds getting
all packages installed:
* DOWNLOAD_ONLY=1 .../build_intiramfs.sh puts all packages in to pkgs
* installation when running w/o DOWNLOAD_ONLY installs all files
matching '*.deb' - which results in all packages being installed
Additionally the TO_DOWNLOAD variable contained all packages anyways
(apt-rdepends yields packagenames - the file-existance check matches
on deb-filenames (containing version+arch) - so this was stripped out
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
switch to our new unified ubuntu-kernels git origin, allows for
simpler update (no buildsys change required anymore)
The two patches dropped are:
- removal of xr-usb-uart, which was patched in by ubuntu (thus no
KConfig switch, and got now removed, see:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1945938
- a fix for a syntax error in Ubuntu patches, which is fixed in newer
repo already
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
as trying to applying them twice won't work, so rather move that out
to its own target, that won't be repeated if new enough.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
as otherwise it's unnecessarily harder to map errors from `patch` to
the respective patch file.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
makes for huge "source" tarballs with DSC, but much nicer than
downloading them on build.
Still missing: using apt-rdepends should be avoided either way..
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
This change makes Busybox symbolically link its inbuilt utils to their
respective locations, allowing them to be invoked through their
absolute path. This means that Busybox inbuilts like e.g. `mount`
and `env` become available at `/bin/mount` and `/usr/bin/env` etc.
In particular, when entering any of the snapshot directories in
`.zfs/snapshot` (e.g. via `cd` or `ls`), ZFS normally mounts the
respective snapshot(s) on demand, using `/usr/bin/env mount [...]`
internally. Before this change, this would fail, rendering ZFS
snapshots unavailable when restoring individual files.
Signed-off-by: Max Carrara <m.carrara@proxmox.com>
use the one from the debian package, as that's what we actually track
but fall back to the cargo one from the init shim (which we normally
don't bump) for developer convenience (env! fails the build if the
variable isn't set).
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
instead of using crates.io - the build still requires network access and
depends on configured APT sources in the build environment, so there's room for
further improvement.
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
the path changed to not include serial8250 directly anymore in jammy
kernel and we just don't really care, not much winning by dying here
if the base ttyS1 dev exists..
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
FILE_LOCKING is required for LVM, the 4 lib* dependencies are not
automatically resolved for some reason, but are required.
Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>