Apparently the perfomance price for compression is to steep to apply it
for all objects >= 64 and < 512 in size, as measured by Arjan Van De
Ven, hence increase the threshold to 512 which yields better results.
We need to tell the X server to grab the keyboards
and mice associated with a hotplugged seat, so that
it doesn't have the ability to control the kernel
vt consoles.
Udev does no longer automatically create udev rules in /etc from the
device hotplug path.
No device name reservation will happen anymore; this model creates
too many problems for setups with many device changes or media which
is booted on different hardware.
Enumerated device names which are based on device discovery order or
on persistent on-disk name reservation will in general not be supported
by udev in the future. It is a problem that can not be solved properly,
and it always creates new problems at the same time it tries to solve
the original one. Udev will no longer pretend it can solve these issues,
and people should switch to available alternatives which provide the
far better compromise.
From now on, udev will only create /dev/cdrom for the first optical
drive, and if the drive is capable /dev/dvd. No other devices will
get any compatibility symlinks or enumerated device names like cdrom1,
cdrom2, and so on. The /dev/cdrom and /dev/dvd links have by default
a negative link priority, which will cause them to be overwritten by
any other device which clains the same names with already existing
udev rules.
If stable device names are needed, the /dev/disk/by-id/ links, which
uniquely identify a specific piece of hardware should be used. The links
usually contain a device serial number and the link names will not depend
on device discovery order.
If completely identical devices with identical or no serial number
need to be handled at the same time, the /dev/disk/by-path/ links can
be used. These links depend on the physical port which is used to connect
the device. It will change when the same device is moved to a different
port or host adapter.
If custom names are needed, custom udev rules which match on specific
device properties need to be added by the administrator.
When systemd starts, plymouth may be already displaying progress
graphically. Do not switch the console to text mode at that time.
All other users of reset_terminal_fd() do the switch as before.
This avoids a graphical glitch with plymouth, especially visible with
vesafb, but could be also seen as a sub-second blink with radeon.
https://bugzilla.redhat.com/show_bug.cgi?id=785548
Device nodes might have been deleted again by the kernel before an
'add' or 'change' event is even started. We need to run all rules,
regardless of the state in /dev.
Tom Gundersen noticed a regression where comment=systemd.automount in
fstab no longer prevented the adding of the After=foo.mount dependency
into local-fs.target. He bisected it to commit 9ddc4a26.
It turns out that clearing the default_dependencies flag is necessary
after all, in order to avoid complementing of Wants= with After= in the
target unit. We still want to add the dependencies on quota units and
umount.target though.
In preparation for https://bugzilla.gnome.org/show_bug.cgi?id=655380 we
decided it's better to include the multi-seat X wrapper in systemd,
rather than gdm. (Side effect: this makes this accessible for other
DMs)
This is a stop-gap for now, until X gins proper multi-seat graphics
support at which point this code will go away without replacement.
When we read the 'uevent' file we need to make sure, that we do not
read the relative DEVNAME= path provided by the kernel and overwrite
the absolute path udev expects here.
Hi,
during the builds for Fedora/s390x I've found that systemd v38 fails to
build on big-endian platforms.
...
make[2]: Entering directory `/root/systemd'
CC src/journal/libsystemd_journal_la-sd-journal.lo
src/journal/sd-journal.c: In function 'init_location':
src/journal/sd-journal.c:69:22: error: incompatible types when
initializing type 'long unsigned int' using type 'sd_id128_t'
src/journal/sd-journal.c:69:20: error: incompatible types when assigning
to type 'sd_id128_t' from type 'long unsigned int'
make[2]: *** [src/journal/libsystemd_journal_la-sd-journal.lo] Error 1
I see the problem in using le64toh() on the 16 bytes boot_id structure
in init_location()
Please see
http://s390.koji.fedoraproject.org/koji/taskinfo?taskID=544375 for a
full build log and attachment for a proposed fix.
With regards
Dan
Albert Strasheim reported a socket unit with Accept=yes was failing
sometimes.
getpeername() returns ENOTCONN if the connection was killed by TCP RST.
The socket unit must not fail when it happens.
Reproducer available at:
https://bugzilla.redhat.com/show_bug.cgi?id=783344
Since the addition of ControlGroupPersistent, systemd is trivially
killed by "systemctl status any.service".
bus_property_append_bool must not be used for a tri-state int.
Also, should it really "b", or do we want the tri-state nature to be seen?
For now just comment out the buggy DBus property.
The pid file watch could outlive the service unit if a daemon-reload
request came at the right time. The inotify event would then be
delivered to who knows where.
Fix it by unwatching in the service destructor.
Further changes will be needed to preserve the state of the pid file
watch across daemon-reload. For now let's just fix the crash observed
by Jóhann Guðmundsson:
Assertion 's->state == SERVICE_START || s->state == SERVICE_START_POST'
failed at src/service.c:2609, function service_fd_event(). Aborting
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=783118
Some broken kernel drivers load firmware synchronously in the module init
path and block modprobe until the firmware request is fulfilled.
The modprobe-generated firmware request is a direct child device of the
device which caused modprobe to run. Child device event are blocked until
the parent device is handled. This dead-locks until the kernel firmware
loading timeout of 60 seconds is reached.
The hanging modprobe event should now time-out and allow the firmware
event to run before the 60 second kernel timeout.
<ridikulus_rat> 60-persistent-storage.rules gpt by-partlabel/by-partuuid
symlinks not created in udev-177 util-linux-2.20.1 kmod-3 in Archlinux x86_64.
<falconindy> ridikulus_rat: fix the rule, or fix the blkid builtin ;)
<kay> oh, i missed the ID_ stuff? :)
The way the various properties[] arrays are initialized is inefficient:
- only the .data members change at runtime, yet the whole arrays of
properties with all the fields are constructed on the stack one by
one by the code.
- there's duplication, eg. the properties of "org.freedesktop.systemd1.Unit"
are repeated in several unit types.
Fix it by moving the information about properties into static const
sections. Instead of storing the .data directly in the property, store
a constant offset from a run-time base.
The small arrays of struct BusBoundProperties bind together the constant
information with the right runtime information (the base pointer).
On my system the code shrinks by 60 KB, data increases by 10 KB.
Now that objects of all unit types are allocated the exact amount of
memory they need, the Unit union has lost its purpose. Remove it.
"Unit" is a more natural name for the base unit class than "Meta", so
rename Meta to Unit.
Access to members of the base class gets simplified.
The storage of the unit objects on the heap is currently not very
efficient. For every unit object we allocate a chunk of memory as large
as the biggest unit type, although there are significant differences in
the units' real requirements.
pahole shows the following sizes of structs:
488 Target
496 Snapshot
512 Device
528 Path
560 Timer
576 Automount
1080 Socket
1160 Swap
1168 Service
1280 Mount
Usually there aren't many targets or snapshots in the system, but Device
is one of the most common unit types and for every one we waste
1280 - 512 = 768 bytes.
Fix it by allocating only the right amount for the given unit type.
On my machine (x86_64, with 39 LVM volumes) this decreases systemd's
USS (unique set size) by more than 300 KB.
quotacheck.service and quotaon.service were not pulled in for fstab mounts.
Fix it by not clearing the default_dependencies flag.
The root filesystem may have quotas too, so don't check for "/" there.
No need to have duplicate code for adding dependencies on umount.target.
https://bugzilla.redhat.com/show_bug.cgi?id=773431
- Include exported package information
- Include C include information
- g_udev_device_get_parent & g_udev_device_get_parent_with_subsystem
transfer ownership of their return values
Signed-off-by: Martin Pitt <martin.pitt@ubuntu.com>
When we merge units that some kind of object points to, those pointers
might become invalidated, and needs to be updated. Introduce a UnitRef
struct which links up all the unit references, to ensure corrected
references.
At the same time, drop configured_sockets in the Service object, and
replace it by proper UNIT_TRIGGERS resp. UNIT_TRIGGERED_BY dependencies,
which allow us to simplify a lot of code.
Use a non-blocking syslog socket if logging from PID 1.
If sendmsg fails with EAGAIN, fall back to kmsg or console only for the
current message. Next message will try syslog again.
Chen Jie observed and analyzed a deadlock. Assuming systemd-kmsg-syslogd
is already stopped, but rsyslogd is not started yet:
1. systemd makes a synchronous call to dbus-daemon.
2. dbus-daemon wants to write something to syslog.
3. syslog needs to be started by systemd.
... but cannot be, because systemd is waiting in 1.
Solve this by avoiding synchronous D-Bus calls. I had to write an async
bus registration call. Interestingly, D-Bus authors anticipated this, in
documentation to dbus_bus_set_unique_name():
> The only reason to use this function is to re-implement the equivalent
> of dbus_bus_register() yourself. One (probably unusual) reason to do
> that might be to do the bus registration call asynchronously instead
> of synchronously.
Lennart's comments from IRC:
> though I think this doesn't fix the problem in its entirety
> simply because dbus_connection_open_private() itself is still synchronous
> i.e. the connect() call behind it is not async
> I think I listed that issue actually on some D-Bus todo list
> i.e. to make dbus_connection_get() fully async
> but that's going to be hard
> so your patch looks good
So it may not be perfect, but it's clearly an improvement.
I did not manage to reproduce the original deadlock with the patch.
