During daemon-reload (or daemon-reexec), when a unit becomes an alias to
another unit, deserialising the alias's stale serialised state can
corrupt the canonical unit's live runtime state.
Consider this scenario:
1. Before reload:
- a.service is running
- b.service was stopped earlier and is dead
- Both exist as independent units
2. User creates an alias to migrate from b -> a:
- `ln -s /run/systemd/system/a.service /etc/systemd/system/b.service`
3. daemon-reload triggers serialisation. State file contains both units:
- a.service -> state=running, cgroup=/system.slice/a.service, PID=1234,
...
- b.service -> state=dead, cgroup=(empty), no PIDs, ...
4. During deserialisation:
- Processes a.service: loads Unit A, deserialises -> state=RUNNING
- Processes b.service: manager_load_unit() detects symlink, returns Unit
A
- unit_deserialize_state(Unit A, ...) overwrites with b's dead state
5. The result is that:
- Unit A incorrectly shows state=dead despite PID 1234 still running
- If a.service has Upholds= dependents, catch-up logic sees a.service
should be running but is dead
- systemd starts a.service again -> PID 5678
- Two instances run: PID 1234 (left-over) and PID 5678 (new)
This bug is deterministic when serialisation orders a.service before
b.service.
The root cause is that manager_deserialize_one_unit() calls
manager_load_unit(name, &u) which resolves aliases via
unit_follow_merge(), returning the canonical Unit object. However, the
code doesn't distinguish between two cases when u->id differs from the
requested name from the state file. In the corruption case, we're
deserialising an alias entry and unit_deserialize_state() blindly
overwrites the canonical unit's fields with stale data from the old,
independent unit. The serialised b.service then overwrites Unit A's
correct live state.
This commit first scans the serialised unit names, then adds a check
after manager_load_unit():
if (!streq(u->id, name) && set_contains(serialized_units, u->id))
...
This detects when the loaded unit's canonical ID (u->id) differs from
the serialised name, indicating the name is now an alias for a different
unit and the canonical unit also has its own serialised state entry.
If the canonical unit does not have its own serialised state entry, we
keep the state entry. That handles cases where the old name is really
just a rename, and thus the old name is the only serialised state for
the unit. In that case there is no bug, because there is no separate
canonical state entry for the stale alias entry to overwrite.
Skipping is safe because:
1. The canonical unit's own state entry will be correctly deserialised
regardless of order. This fix only prevents other stale alias entries
from corrupting it.
2. unit_merge() has already transferred the necessary data. When
b.service became an alias during unit loading, unit_merge() already
migrated dependencies and references to the canonical unit.
3. After merging, the alias doesn't have its own runtime state. The
serialised data represents b.service when it was independent, which is
now obsolete once the canonical unit also has its own serialised entry.
4. All fields are stale. unit_deserialize_state() would overwrite state,
timestamps, cgroup paths, pids, etc. There's no scenario where we want
this data applied on top of the canonical unit's own serialised state.
This fix also correctly handles unit precedence. For example, imagine
this scenario:
1. `b.service` is a valid, running unit defined in `/run`.
2. The sysadmin creates `ln -s .../a.service /etc/.../b.service`.
3. On reload, the new symlink in `/etc` overrides the unit in `/run`.
The new perspective from the manager side is that `b.service` is now an
alias for `a.service`.
In this case, systemd correctly abandons the old b.service unit, because
that's the intended general semantics of unit file precedence. We also
do that in other cases, like when a unit file in /etc/systemd/system/
masks a vendor-supplied unit file in /lib/systemd/system/, or when an
admin uses systemctl mask to explicitly disable a unit.
In all these scenarios, the configuration with the highest precedence
(in /etc/) is treated as the new source of truth. The old unit's
definition is discarded, and its running processes are (correctly)
abandoned. In that respect we are not doing anything new here.
Some may ask why we shouldn't just ignore the symlink if we think this
case will come up. I think there are multiple very strong reasons not to
do so:
1. It violates unit precedence. The unit design is built on a strict
precedence list. When an admin puts any file in /etc, they are
intentionally overriding everything else. If manager_load_unit were to
"ignore" this file based on runtime state, it would break this
fundamental precedent.
2. It makes daemon-reload stateful. daemon-reload is supposed to be a
simple, stateless operation, basically to read the files on disk and
apply the new configuration. But doing this would make daemon-reload
stateful, because we'd have to read the files on disk, but
cross-reference the current runtime state, and... maybe ignore some
files. This is complex and unpredictable.
3. It also completely ignores the user intent. The admin clearly has
tried to replace the old service with an alias. Ignoring their
instruction is the opposite of what they want.
Fixes: https://github.com/systemd/systemd/issues/38817
Fixes: https://github.com/systemd/systemd/issues/37482
Add a --console-transport= option that selects between virtio-serial
(the default, appearing as /dev/hvc0) and a regular serial port
(appearing as /dev/ttyS0 or /dev/ttyAMA0 depending on architecture).
This is primarily useful for testing purposes, for example to test
sd-stub's automatic console= kernel command line parameter handling. It
allows verifying that the guest OS correctly handles serial console
configurations without virtio.
When serial transport is selected, -serial chardev:console is used on
the QEMU command line to connect the chardev to the platform's default
serial device. This cannot be done via the QEMU config file as on some
platforms (e.g. ARM) the serial device is a sysbus device that can only
be connected via serial_hd() which is populated by -serial.
Previously, loop_device_make_internal() always used the block device
directly (via loop_device_open_from_fd()) for whole-device access,
regardless of sector size. This is incorrect when the GPT partition
table was written with a different sector size than the device reports,
as happens with CD-ROM/ISO boot via El Torito: the device has
2048-byte blocks but the GPT uses 512-byte sectors.
Restructure the sector size handling in loop_device_make_internal():
- Move GPT sector size probing (UINT32_MAX case) before the
block-vs-regular-file split so both paths share the same logic and
O_DIRECT handling. Check f_flags instead of loop_flags for O_DIRECT
detection, since we're probing the original fd before any reopening.
- For block devices, get the device sector size and compare it against
the resolved sector_size. Only use the block device directly when
sector sizes match. When they differ (probed GPT mismatch or explicit
sector size request), fall through to create a real loop device with
the correct sector size.
- Default sector_size=0 to the device sector size for block devices
(instead of always 512), so "no preference" correctly matches the
device's sector size.
Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>
While we're at it, rename to test-loop-util.c so it matches its
source file.
We also drop the root check and solely check for CAP_SYS_ADMIN since
that's sufficient to run the tests.
When running in a console mode (interactive, native, or read-only),
propagate the host's $TERM into the VM by adding TERM= and
systemd.tty.term.hvc0= to the kernel command line.
TERM= is picked up by PID 1 and inherited by services on /dev/console
(such as emergency.service). systemd.tty.term.hvc0= is used by services
directly attached to /dev/hvc0 (such as serial-getty@hvc0.service) which
look up $TERM via the systemd.tty.term.<tty> kernel command line
parameter.
While systemd can auto-detect the terminal type via DCS XTGETTCAP, not
all terminal emulators implement this, so explicitly propagating $TERM
provides a more reliable experience. We skip propagation when $TERM is
unset or set to "unknown" (as is the case in GitHub Actions and some
other CI environments).
Previously this was handled by mkosi synthesizing the corresponding
kernel command line parameters externally.
Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>
This enables deriving the minimum size of the `Verity=hash` partition
using the `Verity=` logic when the size of the `Verity=data` partition
is bigger than the `CopyBlocks=` target.
This enables using `Minimize=true` for an "installer image" and later
using sd-repart to install to a system with reserve space for future
updates by specifying `Size{Min,Max}Bytes=` only in the `Verity=data`
partition, without needing to hardcode the corresponding size for the
`Verity=hash` partition.
While not strictly necessary for `Verity=signature` partitions (since
they have a fixed size) there isn't too much reason to not support it,
since then you can still specify `VerityMatchKey=` to indicate that the
partition is logically still part of that group of partitions.
---
Alternative to: https://github.com/systemd/systemd/pull/41156
Fixes https://github.com/systemd/systemd/issues/40995
This is a rebased version of #40936:
- the first few commits are #41003, which is greenlighted for merging
after the release
- then there's #40923
- and some changes on top
The network generator currently supports many of the options described
by dracut.cmdline(7), but not everything.
This commit adds support for the BOOTIF= option (and the related
rd.bootif= option) used in PXE setups.
This is implemented by treating BOOTIF as a special name/placeholder
when used as an interface name, and expecting a MAC address to be set in
the BOOTIF= parameter. The resulting .network file then uses MACAddress=
in the [Match] section, instead of Name=.
terminal_prepare_query() is called from terminal_get_size() which
operates on an explicitly passed fd — typically /dev/console opened
directly by PID 1 via reset_dev_console_fd(), or a service's TTY via
exec_context_apply_tty_size(). Using terminal_is_dumb() here is wrong
because it additionally checks on_tty(), which tests whether stderr is
a tty. PID 1's stderr may not be a tty (e.g. connected to kmsg or the
journal), causing terminal_is_dumb() to return true and skip the ANSI
query even though the fd we're operating on is a perfectly functional
terminal.
Use getenv_terminal_is_dumb() instead, which only checks $TERM, matching
what terminal_reset_ansi_seq() already does.
Also use it in terminal_get_cursor_position(), which also receives fds
to operate on.
terminal_get_size() writes ANSI escape sequences (CSI 18 and DSR
queries) to the output fd to determine terminal dimensions. This is
called during early boot via reset_dev_console_fd() and from service
execution contexts via exec_context_apply_tty_size().
Previously, these writes used loop_write() on a blocking fd, which
could block indefinitely if the terminal is not consuming data — for
example on a serial console with flow control asserted, or a
disconnected terminal. This is the same problem that was solved for
terminal_reset_ansi_seq() in systemd/systemd#32369 by temporarily
setting the fd to non-blocking mode with a write timeout.
Apply the same pattern here: set the output fd to non-blocking in
terminal_get_size() before issuing the queries, and restore blocking
mode afterward. Change the loop_write() calls in
terminal_query_size_by_dsr() and terminal_query_size_by_csi18() to
loop_write_full() with a 100ms timeout so writes fail gracefully
instead of hanging.
Also introduce the CONSOLE_ANSI_SEQUENCE_TIMEOUT_USEC constant for all
timeouts used across all ANSI sequence writes and reads (vt_disallocate(),
terminal_reset_ansi_seq(), and the two size query functions). 333ms
is now used for all timeouts in terminal-util.c.
Also introduce a cleanup function for resetting a fd back to blocking
mode after it was made non-blocking.
Switch runtime virtiofs mount configuration from systemd.mount-extra=
kernel command line parameters to the fstab.extra credential. This
avoids consuming kernel command line space (which is limited) and
matches the approach used by mkosi.
Each mount is added as an fstab entry in the format:
{tag} {destination} virtiofs {ro|rw},x-initrd.mount
If the user already specified a fstab.extra credential via
--set-credential= or --load-credential=, the virtiofs mount entries
are appended to it rather than conflicting.
Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>
Add DISK_TYPE_SCSI_CD to support attaching disk images as CD-ROM
drives, needed for testing El Torito ISO images built by
systemd-repart.
When --image-disk-type=scsi-cd is specified, the image is attached
with media=cdrom and readonly=on on the drive, using scsi-cd as the
device driver on the SCSI bus. This also works for --extra-drive=
with the scsi-cd: prefix.
The QEMU configuration matches the standard OVMF CD-ROM boot setup:
-drive if=none,media=cdrom,format=raw,readonly=on
-device virtio-scsi-pci
-device scsi-cd
When direct kernel booting with scsi-cd, if the kernel command line
contains "rw", append "ro" to override it since CD-ROMs are
read-only.
Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>
During daemon-reload (or daemon-reexec), when a unit becomes an alias to
another unit, deserialising the alias's stale serialised state can
corrupt the canonical unit's live runtime state.
Consider this scenario:
1. Before reload:
- a.service is running
- b.service was stopped earlier and is dead
- Both exist as independent units
2. User creates an alias to migrate from b -> a:
- `ln -s /run/systemd/system/a.service /etc/systemd/system/b.service`
3. daemon-reload triggers serialisation. State file contains both units:
- a.service -> state=running, cgroup=/system.slice/a.service,
PID=1234, ...
- b.service -> state=dead, cgroup=(empty), no PIDs, ...
4. During deserialisation:
- Processes a.service: loads Unit A, deserialises -> state=RUNNING
- Processes b.service: manager_load_unit() detects symlink, returns
Unit A
- unit_deserialize_state(Unit A, ...) overwrites with b's dead state
5. The result is that:
- Unit A incorrectly shows state=dead despite PID 1234 still running
- If a.service has Upholds= dependents, catch-up logic sees
a.service should be running but is dead
- systemd starts a.service again -> PID 5678
- Two instances run: PID 1234 (left-over) and PID 5678 (new)
This bug is deterministic when serialisation orders a.service before
b.service.
The root cause is that manager_deserialize_one_unit() calls
manager_load_unit(name, &u) which resolves aliases via
unit_follow_merge(), returning the canonical Unit object. However, the
code doesn't distinguish between two cases when u->id differs from the
requested name from the state file. In the corruption case, we're
deserialising an alias entry and unit_deserialize_state() blindly
overwrites the canonical unit's fields with stale data from the old,
independent unit. The serialised b.service then overwrites Unit A's
correct live state.
This commit first scans the serialised unit names, then adds a check
after manager_load_unit():
if (!streq(u->id, name) && set_contains(serialized_units, u->id))
...
This detects when the loaded unit's canonical ID (u->id) differs from
the serialised name, indicating the name is now an alias for a different
unit and the canonical unit also has its own serialised state entry.
If the canonical unit does not have its own serialised state entry, we
keep the state entry. That handles cases where the old name is really
just a rename, and thus the old name is the only serialised state for
the unit. In that case there is no bug, because there is no separate
canonical state entry for the stale alias entry to overwrite.
Skipping is safe because:
1. The canonical unit's own state entry will be correctly deserialised
regardless of order. This fix only prevents other stale alias entries
from corrupting it.
2. unit_merge() has already transferred the necessary data. When
b.service became an alias during unit loading, unit_merge() already
migrated dependencies and references to the canonical unit.
3. After merging, the alias doesn't have its own runtime state. The
serialised data represents b.service when it was independent, which
is now obsolete once the canonical unit also has its own serialised
entry.
4. All fields are stale. unit_deserialize_state() would overwrite state,
timestamps, cgroup paths, pids, etc. There's no scenario where we
want this data applied on top of the canonical unit's own serialised
state.
This fix also correctly handles unit precedence. For example, imagine
this scenario:
1. `b.service` is a valid, running unit defined in `/run`.
2. The sysadmin creates `ln -s .../a.service /etc/.../b.service`.
3. On reload, the new symlink in `/etc` overrides the unit in `/run`.
The new perspective from the manager side is that `b.service` is now an
alias for `a.service`.
In this case, systemd correctly abandons the old b.service unit, because
that's the intended general semantics of unit file precedence. We also
do that in other cases, like when a unit file in /etc/systemd/system/
masks a vendor-supplied unit file in /lib/systemd/system/, or when an
admin uses systemctl mask to explicitly disable a unit.
In all these scenarios, the configuration with the highest precedence
(in /etc/) is treated as the new source of truth. The old unit's
definition is discarded, and its running processes are (correctly)
abandoned. In that respect we are not doing anything new here.
Some may ask why we shouldn't just ignore the symlink if we think this
case will come up. I think there are multiple very strong reasons not to
do so:
1. It violates unit precedence. The unit design is built on a strict
precedence list. When an admin puts any file in /etc, they are
intentionally overriding everything else. If manager_load_unit were
to "ignore" this file based on runtime state, it would break this
fundamental precedent.
2. It makes daemon-reload stateful. daemon-reload is supposed to be a
simple, stateless operation, basically to read the files on disk and
apply the new configuration. But doing this would make daemon-reload
stateful, because we'd have to read the files on disk, but
cross-reference the current runtime state, and... maybe ignore some
files. This is complex and unpredictable.
3. It also completely ignores the user intent. The admin clearly has
tried to replace the old service with an alias. Ignoring their
instruction is the opposite of what they want.
Fixes: https://github.com/systemd/systemd/issues/38817
Fixes: https://github.com/systemd/systemd/issues/37482
QEMU's stdio chardev sets O_NONBLOCK on both stdin and stdout (see
chardev/char-stdio.c [1] and chardev/char-fd.c [2]). Since forked
processes share file descriptions, and on a terminal all three stdio
fds typically reference the same file description, this affects our
own stdio too.
Avoid this by using a PTY with chardev serial instead of chardev
stdio for native console mode, matching the approach already used
for interactive and read-only modes. The PTY forwarder shovels bytes
transparently between our stdio and QEMU's PTY using the new
PTY_FORWARD_DUMB_TERMINAL and PTY_FORWARD_TRANSPARENT flags, which
disable terminal decoration (background tinting, window title, OSC
context) and escape sequence handling (Ctrl-] exit, hotkeys)
respectively.
The chardev is configured with mux=on so the QEMU monitor remains
accessible via Ctrl-a c.
Also dedup CONSOLE_NATIVE, CONSOLE_READ_ONLY, and CONSOLE_INTERACTIVE
handling by using fallthrough, with the only differences being the
ptyfwd flags, mux setting, and monitor section.
[1] https://gitlab.com/qemu-project/qemu/-/blob/master/chardev/char-stdio.c
[2] https://gitlab.com/qemu-project/qemu/-/blob/master/chardev/char-fd.c
Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>
Instead of requiring exactly one serial device and assuming ttyS0,
extract the COM port index from the ACPI device path and use the
uart I/O port address format for the console= kernel argument.
On x86, the ACPI UID for PNP0501 (16550 UART) maps directly to the
COM port number: UID 0 = COM1 (0x3F8), UID 1 = COM2 (0x2F8), etc.
The I/O port addresses are fixed in the kernel (see
arch/x86/include/asm/serial.h). Using the console=uart,io,<addr>
format (see Documentation/admin-guide/kernel-parameters.txt) addresses
the UART by I/O port directly rather than relying on ttyS naming,
and also provides early console output before the full serial driver
loads.
Restrict the entire serial console auto-detection to x86. On non-x86
(e.g. ARM with PL011 UARTs), displays may be available without GOP
(e.g. simple-framebuffer via device tree), serial device indices are
assigned dynamically during probe rather than being fixed to I/O port
addresses, and the kernel has its own console auto-detection via DT
stdout-path.
When ConOut has no device path (ConSplitter), all text output handles
are enumerated. If multiple handles have PNP0501 UART nodes with
different UIDs, bail out rather than guessing.
Add ACPI_DP device path subtype, ACPI_HID_DEVICE_PATH struct, and
EISA_PNP_ID() macro to device-path.h for parsing ACPI device path
nodes. Remove MSG_UART_DP, device_path_has_uart(), count_serial_devices()
and proto/serial-io.h (no longer needed).
Move all the console logic to console.c as well.
Add gpt_probe() which probes for a GPT partition table at various sector
sizes (512-4096) and optionally returns the header and partition entries.
Returns the detected sector size on success, 0 if no GPT was found, or
negative errno on error.
Refactor probe_sector_size() in dissect-image.c to be a thin wrapper
around gpt_probe().
Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>
This commit tweaks the TEST-74-AUX-UTILS.varlinkctl.sh code
to use `systemd-notify --fork $UPGRADE_SERVER` instead of
the (ugly) timeout.
This also fixes a stale comment in around
`Test --upgrade with stdin redirected from a regular file`.
Thanks to Daan for suggesting this!
Instead of exiting in exec_with_listen_fds() just return an error
and do the actual _exit() in the caller. Much nicer this way.
Thanks for Lennart for suggesting this.
This commit fixes an issue with the fd handling in
sd_varlink_call_and_upgrade() when one direction of the output
FDs is unset.
Thanks to Lennart for spotting this and suggesting the fix.
Pass -no-user-config while we're at it to avoid loading qemu config
from /etc which is more likely to cause hard to debug issues rather
than do something useful.
Reduce virtiofsd log noise by setting --log-level=error, and drop
the unnecessary mknod capability with --modcaps=-mknod, matching
mkosi's virtiofsd invocation.
Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>
Fix two issues in WRITING_VM_AND_CONTAINER_MANAGERS.md:
1. The Host OS Integration section implied that -M switch and
machinectl shell/login work for VMs, but they currently only
work for containers. Add a note clarifying this limitation.
2. The Guest OS Integration section said "there's only one" VM
integration API (SMBIOS Product UUID), but VM_INTERFACE.md
documents five. Replace the outdated single-API description
with a reference to VM_INTERFACE.md listing all five.
Fixes#40935
Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>
This enables deriving the minimum size of the `Verity=hash` partition
using the `Verity=` logic when the size of the `Verity=data` partition
is bigger than the `CopyBlocks=` target.
This enables using `Minimize=true` for an "installer image" and later
using sd-repart to install to a system with reserve space for future
updates by specifying `Size{Min,Max}Bytes=` only in the `Verity=data`
partition, without needing to hardcode the corresponding size for the
`Verity=hash` partition.
While not strictly necessary for `Verity=signature` partitions (since
they have a fixed size) there isn't too much reason to not support it,
since then you can still specify `VerityMatchKey=` to indicate that
the partition is logically still part of that group of partitions.
We ensure that if one of the hash uses `CopyBlocks=` that the data
partition does so as well. Similarly if the signature partition does it
checks that the hash and data partition do so as well. This is to
minimize the chance of accidental misconfiguration of mixing
`CopyBlocks=auto` and `CopyBlocks=<path>` and of manually populating
partitions while hash/sig partitions are copied from existing sources.
Follow-up for af5126568a ("nspawn: keep backing files for boot_id and
kmsg bind mounts alive").
The backing files for the boot_id and kmsg bind mounts were previously
created in the inner child. However, /run/host/ is remounted read-only
by mount_all() in the inner child (via the MOUNT_IN_USERNS mount table
entry) before setup_boot_id() and setup_kmsg() run, so creating files
there would fail with EROFS.
Fix this by splitting the file creation into separate functions
(setup_boot_id_file() and setup_kmsg_fifo()) that run in the outer
child, where /run/host/ is still writable. The bind mounts onto /proc
remain in the inner child, since procfs is only mounted there.
Also move the backing files from /run/ to /run/host/ and drop the dot
prefix, since /run/host/ is the container-manager-owned namespace and
there is no need to hide these files there. Additionally, apply
userns_lchown() to the created files, matching the convention used by
all other outer child functions that create files in the container.
The collection of facts is entirely unprivileged and has very little to
do with PID1. PID1 is privileged and single-threaded and a point of
contention, so we shouldn't put things in PID1 that don't need to be
there. A separate service can be enabled/disabled/started/stopped at
will, is easy to sandbox, etc. If it turns out to be necessary to
collect some facts through PID1 in the future, we can always add a
smaller facts endpoint to PID1 again.
The main agent doesn't need to read pr-context.json until all
reviews have finished. This should prevent it from passing unnecessary
data from pr-context.json in the prompt to its subagents, which can just
read that file themselves when needed.
Lennart Poettering
Chris Down
Daan De Meyer
Nikolas Kyx
Mike Yuan
Luca Boccassi
Michael Ferrari
Michael Vogt
Kit Dallege
Christian Hesse
Zbigniew Jędrzejewski-Szmek