Commit graph

7 commits

Author SHA1 Message Date
Thomas Haller
8b25221689
initrd: rework parsing of ip method from "ip="
Dracut supports several options for the "ip=" method.

NetworkManager interprets and handles them in a certain way that aims to
give a similar behavior. But as such it maps different settings ("auth6"
and "dhcp6") to exactly the same behavior.

Add _parse_ip_method() function to normalize these keys, and map their
aliases to the keyword that nm-initrd-generator handles. The advantage
is that you see now in _parse_ip_method() which methods are mapped to
the same behavior, and the later (more complex) code only deals with the
normalized kinds.

Also, use the same validation code at all 3 places where IP methods
can appear, that is

  ip=<method>
  ip=<ifname>:<method>[:...]
  ip=<client-ip>:...:<method>[:...]

Also, dracut supports specifying multiple methods and concatenate them
with comma. nm-initrd-generator only did partly, for example,
`ip=dhcp,dhcp6" would have worked, but only because the code failed
to recognize the string and fell back to the default behavior. It would
not have worked as `ip=<ifname>:dhcp,dhcp6[:...]`. Not all combinations
make sense, but some do. So let _parse_ip_method() detect and handle
them. Currently, they mostly map to "auto", but in the future it might
make sense that `ip=dhcp,local6` is a distinct kind.

Try to tighten up the parsing. It's fine to be forgiving and flexible
about what we parse, but bogus values should not silently be
accepted. However, in order to keep previous behavior, `ip=bogus`
and `ip=<client-ip>:...:<bogus-method>[:...]` explicitly map invalid
method to "auto".
2021-07-29 09:23:22 +02:00
Beniamino Galvani
9060c14ccf initrd: support infiniband pkeys
Introduce a new "ib.pkey=<parent>.<pkey>" command line argument to
create a Infiniband partition.

The new connection has IPv4 and IPv6 enabled by default. Unlike for
VLANs, the generator doesn't create a connection for the parent
Infiniband interface.

See also: https://github.com/dracutdevs/dracut/pull/1538

https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/884
2021-07-26 14:56:18 +02:00
Beniamino Galvani
0a18e97345 initrd: set required-timeout for default IPv4 configuration
If the kernel command-line doesn't contain an explict ip=$method,
currently the generator creates connections with both IPv4 and IPv6
set to 'auto', and both allowed to fail.

Since NM is run in configure-and-quit mode in the initrd, NM can get
an IPv4 address or an IPv6 one (or both) depending on which address
family is quicker to complete. This unpredictable behavior is not
present in the legacy module, which always does IPv4 only by default.

Set a required-timeout of 20 seconds for IPv4, so that NM will
preferably get an IPv4, or will fall back to IPv6.

See also: https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues/729
2021-07-05 15:15:45 +02:00
Beniamino Galvani
fa42ba9df2 initrd: rename NMI_WAIT_DEVICE_TIMEOUT_MS to _MSEC 2021-07-05 15:15:44 +02:00
Beniamino Galvani
c21d4ce125 Revert "initrd: set the bootif MAC in existing connection with ifname"
This reverts commit 389575a6b1.

When the command line contains BOOTIF and there is another ip=
argument specifying an interface name, we can follow 2 approaches:

 a) BOOTIF creates a new distinct connection with DHCP
    (the behaviour before the commit)

 b) the connection generated for ip= will be also be bound to the
    BOOTIF MAC (the behavior introduced by the commit)

Restore a) because we can't be sure that the MAC address refers to the
same interface. In that case it's preferable to generate a different
connection.

https://bugzilla.redhat.com/show_bug.cgi?id=1915493#c35
2021-03-26 07:26:55 +01:00
Beniamino Galvani
0f8fe3c76b initrd: fix crash parsing empty rd.znet argument
Ignore a rd.znet argument without subchannels. When using net.ifnames
(the default), subchannels are used to build the interface name, which
is required to match the right connection.

With net.ifnames=0 the interface name is build using a prefix and a
global counter and therefore in theory it is possible to omit
subchannels. However, without subchannels there won't be a udev rule
that renames the interface and so it can't work.

https://bugzilla.redhat.com/show_bug.cgi?id=1931284
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/783
2021-03-17 10:02:09 +01:00
Thomas Haller
9fc1bf90fa
initrd: move "src/core/initrd/" to "src/nm-initrd-generator/" 2021-03-15 17:10:53 +01:00