This reverts commit 9754d56e9b.
It causes a crash in PID1:
Apr 19 13:49:32 lon systemd[1]: Code should not be reached 'Unhandled socket type.'
at src/core/socket.c:684, function instance_from_socket(). Aborting.
Apr 19 13:49:32 lon systemd[1]: Caught <ABRT>, dumped core as pid 336.
Apr 19 13:49:32 lon systemd[1]: Freezing execution.
create_symlink() do not check the return value of unlink(), this may
confuse the user.
Before the unlink() call we check the 'force' argument. If it is not set
we fail with -EEXIST, otherwise we unlink() the file, therefore the next
symlink() should not fail with -EEXIST (do not count races...).
However since callers may not have appropriate privileges to unlink()
the file we lose the -EPERM or any other errno code of unlink(), and
return the -EEXIST of the next symlink(). Fix this by checking unlink()
results.
Before:
$ systemctl --force --root=~/container-03 set-default multi-user.target
Failed to set default target: File exists
After:
$ systemctl --force --root=~/container-03 set-default multi-user.target
Failed to set default target: Permission denied
Currently "systemctl set-default" will fail to change the default target
due to the 'default.target' being a symlink which is always the case.
To work around this, the user must specify the "--force" switch to be
able to overwrite the existing symlink.
This is clearly a regression that was introduced by commit 718db96199
since it worked before without the "--force" switch and the man pages do
not mention that you need to specify it. It is expected that this is a
symlink.
So just explicity set the force flag to make it work again.
https://bugs.freedesktop.org/show_bug.cgi?id=76623
Reported-by: <code@progandy.de>
NixOS uses Unix domain sockets for certain host <-> container
interaction; i.e. the host connects to a socket visible in the
container's directory tree, where the container uses a .socket unit to
spawn the handler program on demand. This worked in systemd 203, but
in 212 fails with "foo.socket failed to queue service startup job
(Maybe the service file is missing or not a template unit?): No data
available".
The reason is that getpeercred() now returns ENODATA if it can't get
the PID of the client, which happens in this case because the client
is not in the same PID namespace. Since getpeercred() is only used to
generate the instance name, this patch simply handles ENODATA by
creating an instance name "<nr>-unknown".
[zj: reorder clauses and remove (unsigned long) casts.]
sec is not set if have_time is false so avoid using it. have_time
was introduced in 9ff09bcb86 but only
the first uses for sec were covered
Found with scan-build
Rather than allocating/freeing memory for each message read, keep a global read buffer
in the rtnl object. Also, rather than using a fixed size, peek at the pending message
header to get the message size and reallocate as necessary.
This means the API can stay the same as for single-part messages by simply passing the head message around. Unrefing
the head of the linked list unrefs the whole list.
The data in the PCI ids file is randomly inconsistent. Many
subvendor model strings just describe the "product" where the
hardware is built into, not the hardware itself. This causes
some "Network Card Model Foo" to show up as "Laptop Model Bar".
Try to make the best out of this mess and concatenate both
strings to describe the hardware.
Previously, AddMatch/RemoveMatch calls where processed exclusively in
the proxy. That's racy however, since subscribing to a signal might not
complete before the signal is sent due to some subsequent method call.
Hence, in order to expose the same ordering guarantees as dbus1 process
the AddMatch/RemoveMatch calls from the proxy, so that they are
dispatched synchronously to all following messages, thus fixing the
race.
Ultimately, we should probabably dissolve the driver entirely into the
proxy, as it is purely a compatibility feature anyway...
The way the kernel namespaces have been implemented breaks assumptions
udev made regarding uevent sequence numbers. Creating devices in a
namespace "steals" uevents and its sequence numbers from the host. It
confuses the "udevadmin settle" logic, which might block until util a
timeout is reached, even when no uevent is pending.
Remove any assumptions about sequence numbers and deprecate libudev's
API exposing these numbers; none of that can reliably be used anymore
when namespaces are involved.
In contrast to a filename-only argument, find_binary() did not
actually check if an path exists, allowing the code to fail later on.
This was OK, but it seems nicer to treat both paths identically.
Also take advantage of path_make_absolute_cwd doing strdup() by itself
if necessary to simplify.
If a persistent timer has no stamp file yet, it behaves just like a normal
timer until it runs for the first time. If the system is always shut down
while the timer is supposed to run, a stamp file is never created and
Peristent=true has no effect.
This patch fixes this by creating a stamp file with the current time
when the timer is first started.
- Negative/positive errno mixup caused duplicates not to be detected properly.
Now we get a warning about some duplicate entries in our own catalogs...
- Errors in update_catalog would be ignored, but they should not be.
