When a new wireless network interface is created by the kernel, it emits
both RTM_NEWLINK and NL80211_CMD_NEW_INTERFACE. These events can arrive
in either order and networkd must behave correctly in both cases.
The typical case is that RTM_NEWLINK is handled first, in which case
networkd creates a Link object and starts tracking it. When the
NL80211_CMD_NEW_INTERFACE message is handled, networkd then populates
the Link object with relevant wireless properties such as wireless
interface type (managed, AP, etc.).
In the event that the order is reversed however, networkd will fail to
populate these wireless properties because at the time of processing the
nl80211 message, the link is considered unknown. In that case, a debug
message is emitted:
systemd-networkd[467]: nl80211: received new_interface(7) message for link '109' we don't know about, ignoring.
This is problematic because after the subsequent RTM_NEWLINK message,
networkd will have an incomplete view of the link. In particular, if a
.network configuration matches on some of the missing wireless
properties, such as WLANInterfaceType=, then it will never match.
The above race can be reproduced by using the mac80211_hwsim driver.
Suppose that there exists a .network configuration:
[Match]
WLANInterfaceType=ap
...
Now loop the creation/destruction of such an AP interface:
while true
do
iw dev wlan0 interface add uap0 type __ap
iw dev uap0 del
done
The above debug message from networkd will then be observed very
quickly. And in that event, the .network file will fail to match.
To address the above race, have the nl80211 message handler store the
interface index in a set in case a Link object is not found on
NL80211_CMD_NEW_INTERFACE. The handler for RTM_NEWLINK can then query
this set, and explicitly request the wireless properties from nl80211
upon the creation of the Link object.
-1 was used everywhere, but -EBADF or -EBADFD started being used in various
places. Let's make things consistent in the new style.
Note that there are two candidates:
EBADF 9 Bad file descriptor
EBADFD 77 File descriptor in bad state
Since we're initializating the fd, we're just assigning a value that means
"no fd yet", so it's just a bad file descriptor, and the first errno fits
better. If instead we had a valid file descriptor that became invalid because
of some operation or state change, the other errno would fit better.
In some places, initialization is dropped if unnecessary.
The name "def.h" originates from before the rule of "no needless abbreviations"
was established. Let's rename the file to clarify that it contains a collection
of various semi-related constants.
I changed imports of util.h to initrd-util.h, or added an import of
initrd-util.h, to keep compilation working. It turns out that many files didn't
import util.h directly.
When viewing the patch, don't be confused by git rename detection logic:
a new .c file is added and two functions moved into it.
Previously, interfaces are partially reconfigured in a spurious way.
Let's use the same way as `networkctl reconfigure`.
Hopefully fixes#14987 and #24997.
We have RxBufferSize= and TxBufferSize= in .link files. Let's use the same
abbreviation here. OTOH, "inc" could be short for "increment" or "increase",
let's avoid that.
When the upstream link gained a lease, then several downstream links may
not appear yet. Previously, the explicitly specified subnet ID for a
downstream link which appears later may be already assigned to an
interface which does not request specific subnet ID.
To avoid such situation, this makes all specified IDs are excluded when
searching free IDs.
As a side effect, we can avoid the second call of
dhcp6_pd_distribute_prefix().
When a system has thousands of downstream interfaces, previously the
total cost of finding free subnet ID was O(n^2), where n is the number
of downstream interfaces.
This makes assigned prefixes are managed by Manager with Hashmap. So,
the cost becomes O(n log n).
This also changes the logic when Priority= is not specified.
Previously, we request without FRA_PRIORITY attribute and kernel picks
the highest unused priority for the rule.
This makes networkd picks the highest unused priority and always request
FRA_PRIORITY attribute.
Previously, when UUID is requested for DUID, then the clients are
configured in callback of bus methods.
But now, 'request queue' was implemented, so we can use it to wait until
the product UUID is obtained.
Previously, link was reconfigured with `link_carrier_reset()`, but it
just re-request to configure static addresses, routes, etc, and restart
engines (e.g. DHCP client).
However, after coming back from sleep, several link information may be
changed, especially the wifi access point may be different. So, we may
need to reconfigure the interface.
Previously, several failures in link_carrier_gained() make link enter
failed state, and other errors are ignored. Now, all failures in
link_carrier_gained(), moreover, link_update() are critical.
networkd already has all information about routes. It is not necessary
to re-read them by using local_gateways().
This also makes manager_find_uplink() take family.
Yu Watanabe
David Tardon
Lennart Poettering
Alvin Šipraga
Zbigniew Jędrzejewski-Szmek
williamsumendap
gtwang01