Creating the rtnl context is cheap, but freeing it may not be, due to
synchronous close().
Also drop some excessive logging. We now log about the changing ifname
exactly once.
Custom endpoints are alternative connection points to a bus, allowing
specific policy to be uploaded.
Add two functions to bus-kernel. One to create such endpoints, and another
one for setting a policy for them.
We must not access slot->floating after we possible dropped the last
reference to it. Fix all callback-invocations to first check
slot->floating and possible disconnect the slot, then release the last
reference.
The bus_map_all_properties() helper calls
org.freedesktop.DBus.Properties.GetAll() on a given target and parses the
result according to a given property-table. This simplifies dealing with
DBus.Properties significantly. However, the function is blocking and thus
not really useful in many situations.
This patch extracts the core of this function and adds two new helpers
which directly take dbus-messages as arguments. This way, you can issue
asynchronous requests and parse the result via these helpers:
bus_message_map_all_properties():
This is the same as bus_map_all_properties() but takes the result
message from a GetAll() request as argument. You can thus issue an
asynchronous GetAll() request and then use this helper once you got
the result.
bus_message_map_properties_changed():
This function takes a signal-message that was retrieved via a
PropertiesChanged signal and then parses it like if you retrieved
it via GetAll(). Furthermore, this function returns the number of
matched properties that got invalidated by the PropertiesChanged
signal, but didn't carry the new value. This way, the caller can
issue a new GetAll() request and then parse the result.
The old function bus_map_all_properties() is functionally unchanged, but
now uses bus_message_map_all_properties() internally.
This will allow sd-event to be integrated into an external event loop, which
in turn will allow (say) glib-based applications to use our various libraries,
without manually integrating each of them (bus, rtnl, dhcp, ...).
The external event-loop should integrate sd-event int he following way:
Every iteration must start with a call to sd_event_prepare(), which will
return 0 if no event sources are ready to be processed, a positive value if
they are and a negative value on error. sd_event_prepare() may only be called
following sd_event_dispatch(); a call to sd_event_wait() indicating that no
sources are ready to be dispatched; or a failed call to sd_event_dispatch() or
sd_event_wait().
A successful call to sd_event_prepare() indicating that no event sources are
ready to be dispatched must be followed by a call to sd_event_wait(),
which will return 0 if it timed out without event sources being ready to
be processed, a negative value on error and a positive value otherwise.
sd_event_wait() may only be called following a successful call to
sd_event_prepare() indicating that no event sources are ready to be dispatched.
If sd_event_wait() indicates that some events sources are ready to be
dispatched, it must be followed by a call to sd_event_dispatch(). This
is the only time sd_event_dispatch() may be called.
This should fix a race where a service thatis idle drops its name, and
is immediately requested by another client, which causes dbus-daemon to
ask systemd to activate it again, but since systemd still assumes it is
running it won't do anything.
Now, that the memfd stuff is not exported anymore, we can simplify a few
things:
Use assert() instead of assert_return(), since this is used internally
only, and we should be less permissive then.
No need to pass an allocated fd back by call-by-reference, we can just
directly return it.
Mapping files as MAP_SHARED is handled by the kernel as 'writable'
mapping. Always! Even with PROT_READ. Reason for that is,
mprotect(PROT_WRITE) could change the mapping underneath and currently
there is no kernel infrastructure to add protection there. This might
change in the future, but until then, map sealed files as MAP_PRIVATE so
we don't get EPERM.
On systems without properly setup systemd, cg_get_root_path returns
-ENOENT. This means that busctl doesn't display much information.
busctl monitor also fails whenever it intercepts messages.
This fix fakes creates a fake "/" root cgroup which lets busctl work
on such systems.
First, let's drop the "bus" argument, we can determine it from the
message anyway.
Secondly, determine the right callback/userdata pair automatically from
what is currently is being dispatched. This should simplify things a lot
for us, since it makes it unnecessary to pass pointers through the
original handlers through all functions when we process messages, which
might require authentication.
Remove the sd_ prefix from internal functions and get rid of the sd_memfd
type. As a memfd is now just a native file descriptor, we can get rid of our
own wrapper type, and also use close() and dup() on them directly.
This is a generalization of the vtable privilege check we already have,
but exported, and hence useful when preparing for a polkit change.
This will deal with the complexity that on dbus1 one cannot trust the
capability field we retrieve via the bus, since it is read via
/proc/$$/stat (and thus might be out-of-date) rather than directly from
the message (like on kdbus) or bus connection (as for uid creds on
dbus1).
Also, port over all code to this new API.
For now this only exposes the domain name (DHCP Option 15), and not
the search string (DHCP Option 119), which will be implemented in
a follow-up patch.
This is not certain to be likely.
Lennart says: a frequent usecase is invoking some function regularly in intervals
in such a case every single iteration we'll have to rearm
i is being used incorrectly. It is used to refer to the number of
indexes calculated so far (out of k). However, it is also incremented
when a new hash key is being used. This means that the results are
inconsistent with the desired behavior described in PORTING-DBUS1
document.
The expected result is that for the default values of m and k (512, 8)
the 1st hash key should produce 4 indexes. The second hash key is used
for the next 4 and overall 8 indexes into m are calculated.
The current behavior results in 6 indexes being calculated, 4 coming
from hash key 1 and 2 others from hash key 5.
Philippe De Swert
Tom Gundersen
Daniel Mack
Ruben Kerkhof
David Herrmann
Hristo Venev
Lennart Poettering
Thomas Hindoe Paaboel Andersen
Denis Kenzior
Stef Walter