The image policy might be configured to enforce verity protection on the
root or usr partitions. Currently, we get around that by ignoring the root
and usr partitions, given that by the time this code runs in the rootfs these
will have already been mounted.
However, we can't actually ignore the root partition, as we will enable
growfs and/or remounting the root partition rw based on the root partition
we find in the image.
Instead, let's just read any roothash= or usrhash= from the kernel command line
and pass in a populated verity settings struct into dissect_loop_device()
instead of NULL. If NULL is passed, any verity partitions in the image are
ignored. If we pass the default verity settings, the wrong verity partitions might
potentially be picked up, so we make sure to pass a fully populated instance. This
should be sufficient to satisfy any configured image policy verity protection checks.
Reverts 6d6538fb3fFixes#38788
This commit cleans up the includes for all the small tools across
the tree.
A few cases of returning EXIT_SUCCESS are replaced with returning
0 to avoid including <stdlib.h>.
Split out of #37344.
We carry a special logic for /usr/ in initrd that it initially gets
mounted to /sysusr/usr/ and later bind mounted to /sysroot/usr/.
When hooking up validatefs it is hence necessary to wait for
the bind mount and validate that, rather than the /sysusr/usr/
established as intermediate transport.
Alternative to #37102
We generally let gpt-auto-generator write to the "late" generator dir,
so that the explicit fstab-generator can write to the "middle" one. And
I think we should leave it that way. But we also want to override the
generic systemd-fsck-root.service service potentially, and if we'd do
that in the "late" generator dir, it would be pointless, since that's
ordered *after* the static version hence would never be taken into
consdiration.
Hence clean this up: keep writing to the late dir for everything,
except for the fsck stuff.
DDIs may contain multiple versions of the same OS, or even multiple
OSes. Hence it makes sense to not just pick the "newest", whatever that
might be, but only partitions associated with specific images, or in a
specific version.
Let's a concept for such filtering: a per-designator glob expression
that can be applied to the partition label string, and can be used for
such filtering.
Usecase: when picking UKI belonging to OS image X in version Y, make
sure we only pick a /usr/ partition belonging to X in version Y, and a
root and home partition belonging to X in any version.
This only adds the basic infrastructure, but doesn't actually expose it
anywhere.
So far, we did not support auto-discovery the /usr/ partition at boot.
Change that: on explicit request (i.e. mount.usr=dissect) automatically
discover the /usr/ partition based on our usual dissection logic.
This introduces root=dissect, which is a lot like root=gpt-auto, but
uses the image dissection logic to find the root partition. It's also
based on the GPT efi var telling us about the root disk, but instead of
just picking some suitable root disk manually it uses the dissection
logic.
This is a big step forward, since it means we can actually make
use of the image policy logic for the root fs too.
Let's turn on validatefs automatically for all auto-discovered
partitions.
Let's add an x-systemd.validatefs option to optionally turn this on for
fstab listed file systems.
At the moment the gpt-auto generator does its things we already
transitioned into the host OS, i.e. the root fs and /usr/ are mounted.
Hence suppress image policy checks for those two partitions.
This actually matters, because the root hash/usr hash is taken into
consideration for the image policy checks, but we don't have that in
gpt-auto and hence would refuse operation claiming policy conflicts
event though we never actually operate on the root fs via the dissection
logic.
The partition enumeration only runs on the main system, and we test that
early, hence no point in repeating this in functions further down the
call chain. But let's keep it in place as assert()s, just in case.
Also, move the top-level in_initrd() into add_mounts(), so that the
tests are nicely encapsulated in the code they protect.
So far our login in gpt-auto-generator when run in the initrd has been
to generate the units that wait for /dev/gpt-auto-root to show up and
mount them only if we have the loader partition EFI variables set. This
is of course not the case for network boots with a UKI kernel, which
means auto-gpt would not work for mounting the rootfs.
What's nasty is that we don't know for sure whether the "rootdisk"
loopback device will shown up eventually, as it needs explicit
configuration by the user via the kernel cmdline, or could be configured
entirely indepdenently. Hence, let's tweak the logic when we wat for
/dev/gpt-auto-root as device to mount: make the gpt auto root logic a
tristate: if root=gpt-auto is specified on the cmdline *definitely*
enable the logic. If root= is specified and set to anyting else,
*definitely* disable the logic. And if root= is not specified check for
the EFI partition vars – as before – to conditionalized things.
Or in other words, you can now boot the same image either via ESP/local
boot or via netboot with a kernelcmdline image like this:
rd.systemd.pull=verify=no,machine,blockdev,bootorigin,raw:rootdisk:image.raw root=gpt-auto rootflags=x-systemd.device-timeout=infinity ip=any
We document that rootflags= is honored for gpt-auto,
but only fstab-generator actually carried the logic for
x-systemd.device-timeout= and .mount-timeout=.
Apply that to gpt-auto too.
Ensure that we always attempt to mount the `ESP` partition to `/boot`
when there is no `XBOOTLDR` partition.
Fixes an issue when booting without a `XBOOTLDR` partition and an empty
root partition, since it would mount the `ESP` partition to `/efi/`
unconditionally causing boot entries to not be under `/boot/` as
recommended by the Boot Loader Specification.
- drop unnecessary SYNTHETIC_ERRNO() when the logger does not propagate
error code,
- drop unnecessary '%m' in error message when the error code is
specified with SYNTHETIC_ERRNO(),
- add missing full stop at the end of log message,
- use RET_GATHER(),
- add missing ", ignoring.",
- upeercase the first letter, etc., etc...
After 8a1326581d,
we always check whether there're mounts under
/boot/ or /efi/ first. Let's relax the check
for fstab_has_node hence, since on initrd-less
systems it might produce wrong results.
Looking for the ESP node is useful to shortcut things but if we're told that
the node is not referenced in fstab that doesn't necessarily mean that ESP is
not mounted via fstab. Indeed the check is not reliable in all cases. Firstly
because it assumes that udev already set the symlinks up. This is not the case
for initrd-less boots. Secondly the devname of the ESP partition can be wrongly
constructed by the dissect code. For example, the approach which consists in
appending "p<partnum>" suffix to construct the partition devname from the disk
devname doesn't work for DM devices.
Hence this patch makes the logic more defensive and do not mount neither ESP
nor XBOOTLDR automatically if any path in paths that starts with /efi or /boot
exists.
I cannot see a reason why we should ignore this error, so let's not. We
use RET_GATHER() on the returns anyway, i.e. collect errors but
continue, so it makes sense to collect this one too.
This call is never called in the initrd, hence we can drop the extra
check, as it is redundant. Let's keep it as an assert() though, as a
form of code-enforced documentation.
Let's say "uki" rather than "stub", since that is just too generic, and
we shouldn't limit us to our own stub anyway, but generally define a
concept of a "measured UKI", which is a UKI that measures its part to
PCR 11.
This is mostly preparation for exposing this check to the user via
ConditionSecurity=.
If the key does not contain '-' or '_', then it is not necessary to use
proc_cmdline_key_streq(), and streq() is sufficient.
This also adds missing assertions about 'key' argument.
Follow-up for 6a488fa7cc
Currently, if an fstab entry for /boot/ exists, we'll skip to try /efi/
instead. However, if it's already using the same device as the discovered
one, we should not duplicate the mount.
Before this commit, we only accept the case when LoaderDevicePartUUID
points to the ESP, while XBOOTLDR is mounted unconditionally.
After this commit, we check if LoaderDevicePartUUID points to either
ESP or XBOOTLDR. If it does, mount both, else nothing gets mounted.
With these settings we intend to turn off timeouts for possibly
interactive/slow commands. The officially documented way to turn off the
time-outs is to setting them to infinity. So far we set them to zero
here though.
This lead to some confusiong, for example #18224. Let's fix this by
uniformly spelling out TimeoutSec=infinity.
This doesn't change behaviour. It just makes our generated files match
what we document, without relying on historic compat support.
Fixes: #18224
I started looking into https://github.com/uapi-group/specifications/issues/35.
BLS says:
> Otherwise [no existing XBOOTLDR partition], if on GPT and an ESP is found and
> it is large enough (let’s say at least 1G) it should be used as $BOOT and
> used as primary location to place boot loader menu resources in.
> It is recommended to mount $BOOT to /boot/, and the ESP to /efi/.
DPS says:
> The ESP used for the current boot is automatically mounted to /efi/ (or
> /boot/ as fallback), unless a different partition is mounted there (possibly
> via /etc/fstab, or because the Extended Boot Loader Partition — see below —
> exists) or the directory is non-empty on the root disk.
I don't think we want to mount the same partition in two places.
If the same partition is not mounted in two places, then the two specs are
contradictory.
The code in gpt-auto-generator implemented the logic from the DPS. It is
modified to implement the logic from BLS.
Effectively:
- if both /boot and /efi are available:
- if both XBOOTLDR and ESP exist:
ESP on /efi, XBOOTLDR on /boot
- if only ESP exists:
ESP on /boot
- if only XBOOTLDR exists:
XBOOTLDR on /boot
- if only /boot is available:
- if XBOOTLDR exists:
XBOOTLDR on /boot
- if only ESP exists:
ESP on /boot
- if only /efi is available:
- if ESP exists:
ESP on /efi
"Available" means that it the mount point is not mounted over and does not
contain files. If the directory doesn't exist, it is also "available" and will
be created later when the mount or automount unit is started.
Thus, the generator attempts to match the partitions and mount points to the
extent possible. In all cases, /boot is the primary place to install kernels.
ESP can be found on /boot or /efi, depending on the situation.
If this patch is merged, I'll submit fixes for BLS and DPS to describe the
same logic.
E.g. in logs on jammy-ppc64el in https://github.com/systemd/systemd/pull/27294:
Apr 16 17:42:50 H systemd-gpt-auto-generator[300]: Failed to dissect partition table of block device /dev/sda: No message of desired type
Apr 16 17:42:50 H (sd-execu[295]: /usr/lib/systemd/system-generators/systemd-gpt-auto-generator failed with exit status 1.
ee0e6e476e made this particular condition not an
error. But for other errnos we want to print a better message too.
dissect_loop_device_and_warn() already does this, but it always prints the
error at error level. We want to suppress some of the errors, so let's make the
print helper public and do the error suppression in the caller.
Daan De Meyer
Jelle van der Waa
Mike Yuan
Antonio Alvarez Feijoo
Lennart Poettering
Zbigniew Jędrzejewski-Szmek
Michael Ferrari
Yu Watanabe
Franck Bui
David Tardon