nvmf / IPv6 related fixes for SLE15-SP5 #254
Conversation
ibft_to_cmdline() formats a static IPv6 address like this (without peer, gateway, hostname): ip="[fd09:9a46:b5c1:1fe:1::10/64]:::::ibft0:none" This means that the "mask" field (4th) is left blank. When this is parsed later by parse-ip-opts.sh, it bails out with the error message "Sorry, automatic calculation of netmask is not yet supported". parse-ip-opts.sh rather expects the prefix in the 4th field: ip="[fd09:9a46:b5c1:1fe:1::10]:::64::ibft0:none" This syntax will be correctly transformed to the command ip addr add fd09:9a46:b5c1:1fe::10/64 dev ibft0 This patch fixes the formatting of the "ip=" line in ibft_to_cmdline().
Assuming a default prefix length of 64 by default if no explicit prefix length is given is wrong and known to cause connectivity problems in some networks. A prefix length of 128 should be assumed in this case. See https://www.rfc-editor.org/rfc/rfc5942 (specifically section 5) https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=684009
For static IPv4 configurations, we use arping to check for duplicate IP addresses. arping requires the af_packet module to work, and if arping fails, the IP address will not be set and booting will fail. af_packet may not be loaded / required in the running system, for example if the system had been booted in an IPv6 configuration, or if it had been manually unloaded. Make sure it's included in initramfs in hostonly mode, too.
This patch reverts commit c603419 ("wait for IPv6 RA if using none/static IPv6 assignment"). It's not generally correct wait for a default route to be established for an interface, or to wait for "proto ra" routes in general. For example, if the system is a router itself, it will receive no RAs. In isolated networks, no gateway may be advertized, either. This is similar in spirit to 76f6566 ("Revert "wait for IPv6 RA if using none/static IPv6 assignment"") Whatever c603419 ("wait for IPv6 RA if using none/static IPv6 assignment") was supposed to achieve, it should be done differently.
In my testing, Note that (in the current code) I haven't installed a scipt in the main "initqueue" which is called every second. |
|
Ok, thanks Martin. |
mwilck
force-pushed
the
SLE15-SP5-nbft-ipv6
branch
from
f8e87c9 to
c6d4d1d
Compare
|
Pushed shfmt fixes. |
mwilck
force-pushed
the
SLE15-SP5-nbft-ipv6
branch
from
c6d4d1d to
637beae
Compare
|
Fixed commit message as remarked by @aafeijoo-suse on dracutdevs/pull/2184 |
Rather than trying to connect immediately from parse-nvmf-boot-connections.sh, move the connect logic to an initqueue script which is called at various stages (settle, online, and timeout). In the timeout case, just try every possible connection. Otherwise, use the existing priority logic. This retry logic is the same that the iscsi module uses. The nvme connect-nbft command will do "the right thing" when the connections specified in the NBFT are already established: Already existing connections will be skipped. Currently nvme-cli will print an error message and exit with error status even if all targets are already connected. This should be handled more gracefully on the nvme-cli side.
mwilck
force-pushed
the
SLE15-SP5-nbft-ipv6
branch
from
637beae to
a414e54
Compare
|
What's holding back this PR? It causes almost boot failure for NVMeoF / IPv6 in static configurations. There's practically no way to recover, the system just halts hard. |
I thought we wanted to wait until the upstream code is accepted, because the basic tech preview feature is already there, but maybe I misunderstood. Do we need to include this in 15-SP5 GM? |
IMO, yes. These are fixes for the code we already have in beta. |
Fixes for IPv6 issues that were found in NVMeoF boot testing.
Checklist
This is equivalent to the update of the usptream PR dracutdevs#2184 that I did today (91e1ee9).