journal_file_read_tail_timestamp() is called unconditionally in
next_beyond_location() for every file on every iteration step,
resulting in O(N x files) volatile mmap reads. For large queries
like 'journalctl -n 1000000' this makes the command unusably slow
(~5 minutes on systems with many journal files).
Rather than suppressing the call entirely (which would make the
inotify path fully load-bearing for cross-boot ordering), rate-limit
it to at most once per second per file. This reduces the overhead
from O(N x files) to O(T x files) where T is the iteration time in
seconds, while still providing periodic refresh as a fallback for
any missed inotify events and keeping cross-boot ordering
reasonably fresh.
Embed a RateLimit struct in JournalFile for this purpose.
Measured improvement on a real system: 5:24 -> 2:39 (-51%) for
'journalctl -n 1000000'.
Fixes: #42521
Regression introduced in: #26355 (34af74946e)
Signed-off-by: dongshengyuan <dongshengyuan@uniontech.com>
When the random seed file doesn't exist but we have good entropy
(seeded_by_efi=true), we attempt to create it. This requires a handle
to the \loader\ directory, which may not exist on systems using
UKI+EFISTUB without systemd-boot installed.
Obtain the directory handle by first trying a read-only open; if that
returns EFI_NOT_FOUND, create the directory. We deliberately avoid
requesting write access on an already-present directory because some
firmware implementations return EFI_INVALID_PARAMETER for a
WRITE|CREATE open on an existing directory — this would be logged at
LOG_ERR and abort seed creation on systems where \loader\ exists but
random-seed does not (the normal systemd-boot layout).
Once a handle to \loader\ is obtained, open the seed file relative to
that handle rather than using the full path from root.
Introduced-by: c0e7046c17 ("boot: log about RO I/O errors at debug level.")
Fixes: #42801
Signed-off-by: dongshengyuan <dongshengyuan@uniontech.com>
varlink_call_and_log() does not hand out a new reference for the reply
object, so the caller should not unref it. The _cleanup_(sd_json_variant_unrefp)
on reply was therefore wrong from the start.
The original TAKE_PTR(reply) was working around this incorrect cleanup
by preventing it from firing, but that left reply's refcount one too
high after sd_json_variant_ref(v) incremented the parent's count.
Fix by dropping _cleanup_(sd_json_variant_unrefp) from the reply
variable declaration entirely, as suggested by Lennart Poettering.
Signed-off-by: dongshengyuan <dongshengyuan@uniontech.com>
When setenv("CREDENTIALS_DIRECTORY") fails, log_error_errno() was
called with the stale return value of exec_context_get_credential_directory()
(which is >= 0 on success) instead of errno.
The %m in the format string correctly expands from libc's errno, so
the human-readable log message was unaffected. However, the structured
journal field ERRNO= received an incorrect value (0 or positive),
making automated log analysis and alerting on this failure unreliable.
Signed-off-by: dongshengyuan <dongshengyuan@uniontech.com>
When using --slice-inherit, setting arg_slice would inadvertently munge
user.slice into the current user slice, usually producing a slice name
like user-1000-user.slice. This is treated as a user-UID slice by
user-.slice.d/10-defaults.conf, resulting in a strange description:
literally "User slice for UID user" (instead of an actual user UID).
Keep arg_slice empty when using --slice-inherit so that we actually
inherit from the relevant slice instead of a munged version.
Fixes: #42601
receive_portable_metadata() reads each item into a stack buffer of
PATH_MAX + NAME_MAX + 2 bytes, passes the full sizeof() as the recv
iovec length, and then NUL-terminates with iov_buffer[n] = 0. recvmsg()
can return n equal to the buffer size, so the terminator is written one
byte past the end.
Grow the buffer by one byte and cap the iovec at sizeof - 1, so a full
record is still received and the trailing NUL always fits, matching the
coredump-receive.c reader.
Both glibc and musl provides dgettext(). Hence it is not necessary to
use libintl.so provided by gettext.
This partially reverts 590e226437,
fully reverts commit e6e65dc261 and
bd19ffd9cb.
Then, introduce minimal libintl.h for musl build, to avoid using
libintl.h by gettext, which is typically installed on musl-based
build systems.
SetRebootToBootLoaderEntry on non-EFI systems wrote the boot loader
entry name to /run/systemd/reboot-boot-to-loader-entry (wrong order),
while the getter and unlink both use the correct path
/run/systemd/reboot-to-boot-loader-entry.
The written value was never read back, silently breaking the feature
on non-EFI systems.
Signed-off-by: dongshengyuan <dongshengyuan@uniontech.com>
After walking a hash chain, the loop exits with p == 0. The error()
call for a tail_hash_offset mismatch passed p as the file offset,
printing 0000000000000000 instead of the actual last data object.
Pass 'last' instead, which holds the offset of the final chain entry.
Signed-off-by: dongshengyuan <dongshengyuan@uniontech.com>
--slice and --slice-inherit are intended to make the new service unit
part of a specific slice. Logind is incompatible with that goal, as a
session of any kind will prompt logind to immediately yoink the new
command from the service unit into a new session scope, which does not
inherit from run0's own slice. The use can still explicitly request a
session with --setenv=XDG_SESSION_CLASS=<class>.
Also make --slice and --slice-inherit conflict with --lightweight and
--area, which depend on logind to be effective.
Add support for two newer NUMA memory policies:
- MPOL_PREFERRED_MANY (Linux 5.15): like MPOL_PREFERRED but accepts
a set of nodes instead of a single node, falling back to all nodes
if preferred nodes cannot satisfy the allocation.
- MPOL_WEIGHTED_INTERLEAVE (Linux 6.9): like MPOL_INTERLEAVE but
distributes pages across nodes proportionally to per-node weights
configured via /sys/kernel/mm/mempolicy/weighted_interleave/.
On kernels that do not support the requested policy, set_mempolicy()
returns EINVAL. We convert EINVAL to EOPNOTSUPP only for the two new
policies (MPOL_PREFERRED_MANY, MPOL_WEIGHTED_INTERLEAVE), so that a
bad NUMAMask= for already-supported policies still fails the service
rather than being silently ignored.
The NUMA subsystem being absent (ENOSYS) continues to be handled
silently at debug level, as before.
Varlink serialization uses json_underscorify() on an owned copy of
the policy name string to convert hyphenated names to the underscore
form declared in the IDL enum, avoiding mutation of the read-only
static string table.
Signed-off-by: dongshengyuan <dongshengyuan@uniontech.com>
I couldn't convince vmspawn to start a VM on a Fedora image I just
downloaded, and it was pretty light on any useful details:
$ build/systemd-vmspawn --image ~/Downloads/Fedora-Server-Guest-Generic-Rawhide-20260627.n.0.x86_64.qcow2 --image-format=qcow2 --bind-ro=/tmp/bar; echo $?
░ Spawning VM Fedora-Server-Guest-Generic-Rawhide-20260627.n.0.x8664.qcow2 on /home/mrc0mmand/Downloads/Fedora-Server-Guest-Generic-Rawhide-20260627.n.0.x86_64.qcow2.
░ Press Ctrl-] three times within 1s to kill VM.
1
Turns out that the unix socket path vmspawn generates for the virtiofsd
socket is too long. Let's relay this information to the user as well to
make debugging this a little less painful:
$ build/systemd-vmspawn --image ~/Downloads/Fedora-Server-Guest-Generic-Rawhide-20260627.n.0.x86_64.qcow2 --image-format=qcow2 --bind-ro=/tmp/bar
░ Spawning VM Fedora-Server-Guest-Generic-Rawhide-20260627.n.0.x8664.qcow2 on /home/mrc0mmand/Downloads/Fedora-Server-Guest-Generic-Rawhide-20260627.n.0.x86_64.qcow2.
░ Press Ctrl-] three times within 1s to kill VM.
Failed to prepare unix socket '/run/user/1000/systemd/vmspawn/Fedora-Server-Guest-Generic-Rawhide-20260627.n.0.x8664.qcow2/sock-9594581dcf598992': File name too long
Add a --firstboot mode that interactively walks the user through enrolling a
passphrase, a recovery key, or a FIDO2 token, with one menu entry per suitable
token currently plugged in (driven by a new fido2_enumerate_devices() helper).
Pressing enter at the top-level menu leaves the volume unchanged; for each
already-enrolled credential type the wizard offers to wipe it as part of the
operation. It populates the same EnrollContext the command line and Varlink
paths use, so the actual enrollment goes through the shared enroll_now() path.
A companion --prompt-suppress= option takes a list of slot types: if a slot of
any listed type already exists, the wizard does nothing and exits successfully.
This lets it be hooked into the boot process while staying quiet once the
system has been set up.
The accompanying systemd-cryptenroll-firstboot.service runs this from the
initrd, after systemd-repart has created the encrypted volume but before we
transition to the host, suppressing itself once a password, recovery key or
FIDO2 token is enrolled. To make that work, determine_default_node() now looks
below /sysroot/ when running in the initrd, since the host file systems aren't
at their final location yet.
While the wizard is active it draws the same installer-style chrome (blue bars
at the top and bottom of the terminal) as systemd-sysinstall, using the shared
prompt_loop_yes_no() helper for its wipe confirmations.
Honours the systemd.firstboot= kernel command line option.
Fixes: #36298
Add a Varlink interface for systemd-cryptenroll, building on the EnrollContext
introduced previously. A single Enroll method covers password, recovery-key and
FIDO2 enrollment; PKCS#11 and TPM2 are not exposed for now (they are not
part of the EnrollMechanism allowlist, so the generic InvalidParameter error
applies). A ListSlots method enumerates the currently enrolled keyslots.
The dispatcher populates the same EnrollContext the command line uses and then
runs the shared enroll_now()/prepare_luks()/wipe_slots() paths, so both
front-ends behave identically. FIDO2 enrollment that requires user presence
reports an imminent touch via a non-terminating "state":"touch" reply when the
caller passes 'more'. Credential material (password, FIDO2 PIN, recovery key)
is handled as sensitive, and key files may be passed either by path or as an
fd index.
The server is allocated root-only plus caller's-own-UID, with the listening
socket created in 0644 mode.
Replaces: #31096
This is preparation for the Varlinkification, as then we want to pass
the password in via IPC instead of prompting the user.
Note that this only adds the field, and applies it, but never actually
sets it. That's for the varlinkification later.
Introduce an EnrollContext structure that carries everything the enrollment
and unlocking helpers need, and route all enroll_*()/load_volume_key_*()/
wipe_slots() calls through it. The command line still populates the existing
arg_* globals as before; once parsing is complete they are copied into a
self-contained EnrollContext (which owns its strings/arrays) and the rest of
the code only ever reads from the context.
This is preparation for the upcoming varlinkification of systemd-cryptenroll:
a Varlink dispatcher (and later an interactive first-boot wizard) can populate
the very same EnrollContext without going through the arg_* parsing layer.
To support non-interactive (e.g. Varlink) callers, the context carries an
'interactive' flag: when false, every credential prompt is disabled and the
helpers fail with -ENOPKG (the established "querying disabled via headless"
code) instead of blocking on a tty. Passwords, FIDO2 PINs and PKCS#11 PINs are
all covered, and an optional FIDO2 PIN can be supplied directly via the
context. enroll_recovery() additionally grows a quiet mode that returns the
recovery key instead of printing it.
This also adds one new field to EnrollContext which didn't exist before:
the unlock_password is useful for the Varlink hookup later.
No change in command line behaviour.
Factor a yes/no variant of prompt_loop() out into prompt-util.[ch], so the
various interactive tools can share a single implementation, and convert
systemd-sysinstall's installation confirmation question over to it.
With this change, gcrypt dependency is not mandatory. Hence, allow to build
systemd even when -D gcrypt=enabled but gcrypt devel package is not installed.
This introduce a vtable for journal tagging feature in sd-journal,
and makes libgcrypt dependent features loaded by users (journald,
journalctl, journal-remote, and unit tests) when necessary.
Then, move several components for journal tagging in JournalFile
to JournalAuthContext.
This also introduces wrapper functions that checks gcrypt support.
Luca Boccassi
dongshengyuan
Ronan Pigott
Syed Mohammed Nayyar
Yu Watanabe
Lennart Poettering
Jouke Witteveen
Frantisek Sumsal