2
0
mirror of https://github.com/xcat2/xcat-core.git synced 2026-11-19 17:02:15 +00:00
Commit Graph

29331 Commits

Author SHA1 Message Date
Daniel Hilst cd67199463 fix(netboot): an IPv6-only diskless node boots with no address
xcatroot writes the netplan of the booted system from the interface it found,
and it always wrote "dhcp4: true". An IPv6-only node has no DHCPv4 server, so it
reached its login prompt with no address: Ubuntu 24.04.5 reported
"[FAILED] Failed to start systemd-networkd-w... Wait for Network to be
Configured." and "[FAILED] Failed to start xcatpostinit.service", and the node
answered no ssh.

The address is already on the kernel command line, in the static form dracut
takes. xcat_static_ip6_fields reads it and the netplan carries the address, the
prefix and a default route through the gateway. A node whose command line holds
an IPv4 form keeps dhcp4, because the fields are read only from a bracketed
literal.

Captured by xCAT-test/bats/ipv6_xcatroot_netplan.bats. Reading the fields by
cutting on the first colon, which is what an IPv6 literal breaks, fails the
three parsing assertions and leaves the four refusals passing.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 16:01:34 -03:00
Daniel Hilst 77eaf4d53c fix(ipv6): a diskless node has an address and no route
mknetboot gave an IPv6-only node ip=either6, which asks dracut for DHCPv6.
DHCPv6 assigns one address with a prefix length of 128 and carries no route, and
an xCAT provision network sends no router advertisement: radvd is not installed
and net.ipv6.conf.<nic>.forwarding is 0. The node came up with the address and
no way to reach the server, and the dracut hook reported
"Connecting to [2001:db8:147::1]:80... failed: Network is unreachable."

debian.pm now builds the static form from the address, gateway and prefix the
node already has in the database, the way sles.pm passes hostip and gateway to
linuxrc. either6 remains the answer when the address cannot be resolved.

Captured by xCAT-test/unit/dracut_static_ip6.t. Removing the prefix length from
the parameter fails four of its nine assertions and leaves the bracketing and
the refusal cases passing.

Measured on an IPv6-only Ubuntu 24.04 node: Kea records prefix_len=128 for its
lease, and dracut logged BOUND6 for eth0 before the fetch failed.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 15:27:53 -03:00
Daniel Hilst ddaad73ab4 fix(grub2): a ppc node is given linuxefi when its image carries it
grub2_kernel_commands read the image and not the architecture, so a ppc node
whose grub2 image carries linuxefi was given it. linuxefi loads an EFI binary
and a ppc loader is not one. The previous commit removed the architecture test
along with the defect it carried.

The command now needs both: the architecture whose loader is an EFI binary, and
an image that carries the command.

grub2_netboot_protocol.t covers the three architectures and caught this. Its
loader fixture now records the commands it carries, because the configuration
follows the image.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 15:27:53 -03:00
Daniel Hilst 075e6801e9 fix(grub2): an x86 node stops at the menu with "can't find command `linuxefi'"
setstate in grub2.pm appends "efi" to the kernel and initrd commands for every
node whose architecture matches x86, so the configuration reads linuxefi and
initrdefi. Both come from a downstream patch. grub2-xcat 2.12-2 for Ubuntu is
built from upstream grub and carries neither, so the node loads the loader and
its own configuration over TFTP, prints "can't find command `linuxefi'", asks
for no kernel, and waits at the menu until it is powered off.

The pair now comes from the image served to the node. BootUtils reads the
image, which records the name of every command its modules register, and
answers linuxefi and initrdefi only when the image carries them. An image that
cannot be read is answered with linux and initrd, which every grub2 image
carries.

This reaches a node only when it netboots with grub2. An IPv6 node has no other
loader, because bootfile-url is written for grub2 alone.

Measured on an IPv6-only Ubuntu 24.04 node: with linuxefi the node made no TFTP
request for the kernel; with linux it loaded the kernel and the initial ramdisk
and printed no error.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 14:45:34 -03:00
Daniel Hilst 557c3689f7 test(grub2): a grub2 image without linuxefi is still offered linuxefi
grub2.pm chooses the kernel command from the architecture of the node, so every
x86 node is given linuxefi and initrdefi. Those two come from a downstream patch
and an image built from upstream grub carries neither.

These assertions fail without the fix that follows: the upstream image, the
missing image and the undefined image are all answered with linuxefi.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 14:45:33 -03:00
Daniel Hilst 2f6f029382 fix(xcat-core): a diskless node has no server in its imgurl
mknetboot in debian.pm builds the imgurl from the image server, which is the
'!myipfn!' placeholder whenever noderes.tftpserver and noderes.xcatmaster are
unset. Commit 7ab362835 wrapped that value in format_host_port so an IPv6
server would be bracketed. format_host_port refuses the placeholder, because
format_uri_host validates it as a host name, so the imgurl lost its server and
its port and read "imgurl=http:////install/netboot/.../rootimg.cpio.gz". The
node then had no address to fetch its root image from. This reached every
Ubuntu diskless node, not only an IPv6 one: upstream interpolated the value
directly and the placeholder survived to be substituted.

format_uri_host now returns the placeholder unchanged, the way uri_authority
already passes it through for the XCAT= parameter. The boot loader plugins
substitute it afterwards, and substitute_facing_address brackets it when the
facing address is IPv6.

Measured on an IPv6-only Ubuntu 24.04 node: the grub2 command line now carries
imgurl=http://[2001:db8:147::1]:80/install/... beside ip=either6.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 13:49:07 -03:00
Daniel Hilst 5760b73d57 test(networkutils): the facing-address placeholder is dropped by format_host_port
format_host_port validates its host with format_uri_host, which refuses
'!myipfn!' because it is not a host name. The call returns undef, so a command
line built from it loses both the server and the port.

These assertions fail without the fix that follows: format_host_port and
format_uri_host return nothing for the placeholder, and the imgurl built from it
reads "imgurl=http:///install/..." for an IPv6 and an IPv4 facing address alike.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 13:49:07 -03:00
Daniel Hilst 1c8094e305 fix(template): the IPv6 search list makes YaST refuse the whole profile
The search list added in the previous commit carried no config:type, and the
AutoYaST schema gives searchlist the LIST type, the way it gives it to
nameservers. xmllint against profile.rng of Leap 15.6 reports only "Element
profile failed to validate content", and YaST refuses the profile entirely, so
the node cannot install at all.

searchlist is now emitted as a typed list. Measured against
/usr/share/YaST2/schema/autoyast/rng/profile.rng on a live management node: the
generated profile fails to validate, the same profile with the attribute added
validates, and removing hostname as well changes nothing, so hostname was not
the element at fault.

Captured by xCAT-test/unit/autoyast_ipv6_network.t, which gained an assertion on
the attribute.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 12:48:21 -03:00
Daniel Hilst 278b582cf5 fix(template): an IPv6-only SUSE node is installed as "localhost"
_autoyast_network6 sets dhcp_hostname to false, because an IPv6-only node has no
DHCPv4 to take a hostname from, and it writes no hostname of its own. YaST then
has nothing to set: y2log of an installed node records Y2Network::Hostname with
@static=nil, @transient=nil and @dhcp_hostname=:none, and the node answers
"localhost". The two sibling builders in the same file write both keys.

The dns section now carries the node name and a search list of the node domain.
getNodeNetworkCfg6 already returns the name, so no new lookup is needed.

Captured by xCAT-test/unit/autoyast_ipv6_network.t. Without the two keys its
assertions 17 and 18 fail and the other 16 pass.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 12:26:20 -03:00
Daniel Hilst 0ac354f931 fix(sle): an IPv6 install server empties the AutoYaST profile
pre.sle substitutes the install server for XCATNEXTSERVERHOOK in the media_url
that xCAT::Template writes for every install source. The substitution read
sed -e 's!XCATNEXTSERVERHOOK!'$nextserver'!', with the value unquoted, and
shopt -s nullglob is set earlier in the same script. An IPv6 server is bracketed,
so "[2001:db8::1]" is a glob that matches no file and the whole sed script word
is removed. sed then reads the profile as its script and writes nothing, so
/tmp/profile/modified.xml is empty. AutoYaST falls back to the unmodified profile
and stops on "Disk 'XCATPARTITIONHOOK' was not found", because the pre-script
also substitutes the install disk into that profile.

The substitution moves into substitute_nextserver, which quotes the value. An
IPv4 server was never affected, because it carries no bracket.

Captured by xCAT-test/bats/ipv6_pre_sle_nextserver.bats. With the quotes removed
again, assertions 9 and 10 fail and the IPv4 assertion 11 still passes.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 11:53:52 -03:00
Daniel Hilst f094586706 fix(template): the AutoYaST profile gives an IPv6-only node an IPv4 resolver
_autoyast_network6 passed the node nameserver list through unchanged, so an IPv6-only
node was configured with the IPv4 resolver that site.nameservers names for the cluster.
The node cannot reach it, and an unreachable server in resolv.conf costs every lookup
a timeout.

The list now keeps only the IPv6 servers. When that leaves none, the gateway stands
in, which is the management node and already answers for the node. The <xcatmaster>
substitution is unchanged.

Captured by xCAT-test/unit/autoyast_ipv6_network.t, where two of its sixteen
assertions fail without this change.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 09:58:17 -03:00
Daniel Hilst 7fd7ee7d92 test(template): an IPv6-only node is given an IPv4 resolver it cannot reach
site.nameservers holds one list for the whole cluster, so an IPv6-only node takes the
IPv4 resolver of an IPv4 network when its own network names none. The AutoYaST profile
of a Leap 15.6 node carried <nameserver>192.168.156.1</nameserver> while the node had
no IPv4 address at all.

An unreachable server in resolv.conf is not a failure the node reports. Every lookup
waits for its timeout before the next server is tried.

The test adds a node whose configured resolvers are IPv4 only and asserts the section
carries none of them, and that the gateway stands in.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 09:58:17 -03:00
Daniel Hilst 8eb12a1392 fix(template): an IPv6-only node has its static address rewritten to DHCP
An IPv6-only openSUSE node completed its install with the static IPv6 address in its
AutoYaST profile, then booted with a link-local address only. The applied profile on
the node carried bootproto static and the address, and ifcfg-eth0 read BOOTPROTO=dhcp
with a modification time later than the install.

post.sles.common writes that file. It keeps the interface configuration only when
MANAGEDADDRESSMODE reads static, and managed_address_mode returned
site.managedaddressmode, which was unset. xCAT wrote the static configuration and then
removed it.

managed_address_mode now takes the node and reports static for a node that
xCAT::NetworkUtils::node_is_ipv6_only reports as IPv6-only. A node with any IPv4
address keeps the site value, so an IPv4 and a dual-stack cluster are unchanged.

Captured by xCAT-test/unit/template_ipv6_rendering.t, where one of the three new
assertions fails without this change.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 09:56:37 -03:00
Daniel Hilst 2a6d80fb8c test(template): the address mode is answered for the cluster, not for the node
post.sles.common reads MANAGEDADDRESSMODE and, for anything but static, removes the
interface configuration and writes a new one with BOOTPROTO=dhcp. managed_address_mode
returns site.managedaddressmode, which describes the cluster.

An IPv6-only node therefore has the static IPv6 configuration in its AutoYaST profile
rewritten to DHCP during the post phase, and comes up with a link-local address only.
The node has no IPv4 address to ask a DHCP server for, so the site attribute cannot
answer for it.

The test renders MANAGEDADDRESSMODE for three nodes with the site attribute set to
dhcp: an IPv6-only node, an IPv4 node and a dual-stack node. Only the IPv6-only node
is expected to differ, so the other two hold the site behaviour down.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 09:56:37 -03:00
Daniel Hilst 582fd157f2 fix(ipv6): an IPv6-only Ubuntu diskless node gets a DHCPv4 initrd
The Ubuntu diskless kernel command line carries no dracut ip= parameter. dracut then
configures the interface it needs with DHCPv4, so an IPv6-only node comes up with no
global address, cannot fetch imgurl and never mounts a root.

gen_net_boot_params now returns ip6 beside the ip it already derived, spelled either6,
which asks dracut for DHCPv6 and falls back to router advertisement. debian.pm appends
it for a node that xCAT::NetworkUtils::node_is_ipv6_only reports as IPv6-only, so an
IPv4 node is told exactly what it was told before. The ip the EL path reads is
unchanged.

Captured by xCAT-test/unit/networkutils_dracut_ipv6_boot_param.t. This closes the
derivation; the Ubuntu diskless path is not yet proven end to end on a cluster.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 09:09:23 -03:00
Daniel Hilst 918ed688de test(ipv6): nothing gives the dracut initrd an IPv6 network parameter
An Ubuntu diskless node brings its interface up from the dracut ip= parameter.
debian.pm appends ifname, netdev and BOOTIF and no ip= at all, and dracut configures
the interface it needs with DHCPv4 when the command line carries none. An IPv6-only
node therefore gets no address, cannot fetch imgurl and never mounts a root.

gen_net_boot_params already derives the interface in xCAT's own order and returns the
IPv4 spelling of the parameter. The test asserts the IPv6 spelling beside it for every
branch that names an interface, and asserts the IPv4 spelling is unchanged, because
anaconda.pm is its only caller today.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 09:09:23 -03:00
Daniel Hilst d73343b198 fix(packaging): the openSUSE IPv6 module gap needs a requires, not a hard dependency
The previous commit required Socket6 and IO::Socket::INET6 on every RPM family. That
is wrong for EL and openEuler, where dnf installs a recommended package and the
existing Recommends already delivers both modules.

The gap is openSUSE only, and it has two parts. xCAT-server recommends both modules
inside a guard that names fedora, rhel and openEuler and omits suse_version, so an
openSUSE build declares neither a requires nor a recommends. perl-xCAT declares nothing
at all, although xCAT::SLP, xCAT::Client and xCAT::NetworkUtils all reach IPv6 through
these modules. A recommendation would not close it either: the openSUSE Minimal-VM
image sets solver.onlyRequires = true, so zypper installs no recommended package, which
is why xCAT already names a DHCP server directly.

Both specs now require both modules on openSUSE and recommend them on the families
whose package manager honours a weak dependency. The xCAT-server deb returns to a
Recommends.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 09:05:15 -03:00
Daniel Hilst edea395c29 test(packaging): the IPv6 module dependency is asserted for the wrong families
The assertions added with this test required Socket6 and IO::Socket::INET6 on every
RPM family, which contradicts xcat_server_inet6_optional_dependency.t and the reason
it records: every call site loads the modules with an eval and keeps working on IPv4,
so a hard dependency states a dependency the code does not have.

The measurement that prompted the test still holds, and it is narrower than it looked.
Both modules are absent on a Leap 15.6 management node because the openSUSE Minimal-VM
image sets solver.onlyRequires = true in /etc/zypp/zypp.conf, so zypper installs no
recommended package. dnf installs one, so EL and openEuler need no hard dependency.

The test now reads requires and recommends separately per family: openSUSE requires
both and recommends neither, EL and openEuler recommend both and require neither. The
deb control files are read per package, since perl-xCAT has carried a hard dependency
since before this rule. xcat_server_inet6_optional_dependency.t gains a pointer to the
openSUSE exception so the two files do not disagree in prose.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 09:05:15 -03:00
Daniel Hilst 174668fb60 fix(sles): an IPv6 install server reaches the media_url as "[2001"
The nextserver derivation in pre.sle and pre.sles split the URL authority at the
first colon, which splits an IPv6 literal. YaST then had no usable install source
on a SLE15 or SLES12 node installed over IPv6.

Both scripts now read the authority with the bracket-aware form pre.rhels10 already
uses, and put the brackets back when the host is an IPv6 address, because the value
goes into a media_url. A portless IPv4 URL no longer keeps its path.

Captured by xCAT-test/bats/ipv6_pre_sle_nextserver.bats, which fails 5 of its 8
assertions without this change.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 08:56:37 -03:00
Daniel Hilst c86de4f7a9 test(sles): the SUSE installer cuts an IPv6 install server at its first colon
pre.sle and pre.sles read the install server out of the kernel command line and put
it back into a URL. pre.sle substitutes it for XCATNEXTSERVERHOOK in the media_url of
every install source, and pre.sles substitutes it into #INSTALL_SOURCES_IN_PRE#.

The derivation cut the authority at the first colon. An IPv6 install server written
"autoyast=http://[2001:db8::1]:80/install/autoinst/node01" yielded "[2001", so the
media_url read "http://[2001:80/install/..." and YaST had no install source. A
portless IPv4 URL kept its path as well, and worked only because xCAT always writes
a port.

The test runs the derivation taken from each script against a command line of its
own, so a change to the derivation is measured here.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 08:56:37 -03:00
Daniel Hilst bf3a2f066b fix(packaging): SLP discovery and IPMI over IPv6 are off on an openSUSE node
Socket6 and IO::Socket::INET6 are absent on a Leap 15.6 management node that carries
every other xCAT dependency, so $ip6support in xCAT::SLP and $doipv6 in xCAT::IPMI
evaluate to 0. SLP discovery and IPMI over IPv6 then do nothing and report no error.

Both packages exist: perl-Socket6 and perl-IO-Socket-INET6 are in the openSUSE main
repository and in the EL appstream repository. Nothing asked for them. perl-generators
reads a use statement, and every caller uses eval { require ... }, so the EL build
derives no perl(Socket6) either. xCAT-server named the deb packages as Recommends,
which apt 3.0 solver3 drops when it cannot place them.

perl-xCAT and xCAT-server now require both modules on every RPM family, and the
xCAT-server deb moves them from Recommends to Depends.

Captured by xCAT-test/unit/ipv6_socket_modules_declared.t, which fails 16 of its 20
assertions without this change.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 08:52:27 -03:00
Daniel Hilst 7703372cc2 test(packaging): nothing declares the two Perl modules every IPv6 path needs
xCAT::SLP, xCAT::IPMI, xCAT::Client, xCAT::NetworkUtils and the dhcp, nodestat and
credentials plugins reach IPv6 only through Socket6 and IO::Socket::INET6. Each one
loads them with eval { require ... }, so a missing module is not an error. The guard
variable stays false and the IPv6 branch is never taken.

No family declares the dependency. perl-generators reads a use statement, not a
require inside an eval, so the EL build derives no perl(Socket6) either, and
xCAT-server lists the two deb packages as Recommends, which apt 3.0 solver3 can drop.

The test asserts both specs require both modules on openSUSE, EL and openEuler, and
that both deb control files name them in Depends.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-29 08:52:19 -03:00
Daniel Hilst 2bb169e04f fix(template): an IPv6 node installs and then boots with no address
_autoyast_network6 builds a networking section with a static interface, but did
not set keep_install_network. AutoYaST then keeps the network the installer
used and ignores the interface in the profile.

The node installed over IPv6 and booted with ifcfg-eth0 BOOTPROTO=dhcp and no
address but a link-local one. xcat.sh could not reach the management node:
"logger: failed to connect to 2001:db8:156::1 port syslog-conn" and
"connect: Network is unreachable", so no postscript ran and the node never
registered.

The section now sets keep_install_network to false, which is what makes AutoYaST
write the static interface.

Measured on the installed node: /etc/sysconfig/network/ifcfg-eth0 carried
BOOTPROTO=dhcp and the only address was fe80::4078:c0ff:fea8:9c0a.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 14:48:37 -03:00
Daniel Hilst d1cafdddc3 test(template): the IPv6 profile does not replace the installer network
AutoYaST keeps the network the installer used unless keep_install_network says
otherwise, and the sections _autoyast_network6 replaces set it true. The static
interface in the generated section was therefore ignored.

The test asserts the section sets keep_install_network to false.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 14:48:37 -03:00
Daniel Hilst 29413b5b0e fix(template): an IPv6 node gets an AutoYaST profile that YaST refuses
_autoyast_network6 emitted dhcp_resolv in the dns block of the networking
section. The AutoYaST schema of Leap 15.6 does not accept that element, and one
element it refuses invalidates the whole profile: YaST reported "Extra element
networking in interleave", then "A profile for this machine could not be found
or retrieved", and the installation stopped.

dhcp_resolv says whether DHCP may overwrite resolv.conf. The section configures
the node statically, so there is no DHCP to keep away from resolv.conf.

Measured with the schema the installer ships, /usr/share/YaST2/schema/autoyast/
rng/profile.rng: the profile failed to validate at the networking element, and
removing dhcp_resolv alone made it validate.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 13:28:53 -03:00
Daniel Hilst ebefdec9f8 test(template): the IPv6 networking section carries an element the schema refuses
AutoYaST validated the generated profile and reported "Extra element networking
in interleave", then told the operator no profile could be retrieved. The
element is dhcp_resolv, which the schema of Leap 15.6 does not accept.

The test asserts the section has no dhcp_resolv and keeps dhcp_hostname, which
the schema does accept.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 13:28:53 -03:00
Daniel Hilst 9f5d9be84f fix(sles): the installer of an IPv6-only node configures IPv4 and reaches nothing
sles.pm gave linuxrc no network arguments beyond the net device. linuxrc
configures IPv4 by default, so an IPv6-only node sent an IPv4 DHCPDISCOVER,
received "no/incomplete answer", and reported "Failed to download AutoYaST
file" without ever contacting the install source.

static_ipv6_boot_args() adds ipv6only=1 with the address, prefix, gateway and
IPv6 nameserver of the node. The node is configured statically for the reason
anaconda.pm configures EL statically: the firmware answers DHCPv6 with the
RFC 6355 DUID-UUID that the reservation is keyed on, while the installer builds
a DUID of its own that matches no reservation.

Measured on a Leap 15.6 node: with these arguments the installer fetched
control.xml from http://[2001:db8:156::1]:80/install/leap15.6/x86_64/1 from its
own address 2001:db8:156::10. Without ipv6only=1 linuxrc ignores the rest and
stays on IPv4.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 13:24:43 -03:00
Daniel Hilst b3cb867b87 test(sles): an IPv6-only node gets no installer network configuration
sles.pm put only the net device on the command line, so linuxrc configured IPv4
by default. An IPv6-only node sent an IPv4 DHCPDISCOVER, which the management
node answered "no free leases".

The test covers the arguments for a node with a gateway and an IPv6 nameserver,
for one with neither, that an IPv4 nameserver is left out, and that a node with
no IPv6 configuration gets no arguments.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 13:24:43 -03:00
Daniel Hilst 60d0176d14 fix(dhcp): no range6 is written for an IPv6 subnet that has a dynamic range
addnet6 read the dynamic range from $netcfgs{$net}, where $net is the subnet it
writes, "2001:db8:156::/64". addrangedetection stores the range under
networks.net, "2001:db8:156::", so the lookup always missed and makedhcp warned
"No dynamic range specified" for a network whose row declares one.

The subnet then carried no range6, and a DHCPv6 client whose DUID matches no
reservation received no address. The SUSE installer is such a client: linuxrc
builds its own DUID rather than the DUID-UUID the firmware sends, so it could
not get an address even though the node has a reservation for the same MAC.

dynamic_range_for() looks the range up under the subnet as written and then
under the network address alone.

xCAT-test/unit/dhcp_dhcpv6_conf_file.t fails its prefix-form assertion when the
second lookup is removed.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 12:40:44 -03:00
Daniel Hilst 19779fa2f8 test(dhcp): an IPv6 subnet never gets its dynamic range
addrangedetection keys the dynamic ranges on networks.net, which carries no
prefix length. addnet6 names a subnet "<net>/<prefix>" and looked the range up
under that, so the lookup missed every IPv6 subnet.

The test drives the lookup with both forms of the subnet and holds down that an
IPv4 subnet is unchanged.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 12:40:44 -03:00
Daniel Hilst 837509f8cd fix(xcattest): a semicolon in a conf value truncates the chdef it builds
setup_env_by_configure_file built "chdef -t $type -o $name $attr=$value" and
runcmd passes that to bash. A value containing shell syntax was read as syntax.

vmothersetting separates its settings with a semicolon, so
"vmothersetting=cpumode:host-passthrough;firmware:efi-insecure" ended the chdef
at the semicolon. The node kept cpumode alone, mkvm defined the compute node
with no firmware, and it came up as a BIOS guest. A BIOS guest cannot network
boot over IPv6 and cannot load an EFI grub2, so the node took its DHCPv6 address
and never fetched a loader.

object_definition_command() and shell_quote() build the command with every value
quoted, and the chtab path for [Table_*] quotes its values the same way.

xCAT-test/unit/xcattest_object_quoting.t fails when the quoting is removed.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 12:13:02 -03:00
Daniel Hilst de6c13471f test(xcattest): a conf value that carries shell syntax is not quoted
xcattest applies the [Object_*] sections of a conf by joining each attribute to
a chdef command and running it through bash. Nothing quoted the values, and a
conf carries values a shell reads as syntax.

The test builds the command for a vmothersetting that holds two settings and
asserts the semicolon stays inside one command, that every value is quoted, and
that bash passes the whole value through as one argument.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 12:13:02 -03:00
Daniel Hilst 84f18d5fd5 test(dhcp): nothing asserts the two backends agree on the boot file URL
dhcp6_bootfile_url_for_node builds the RFC 5970 option 59 URL for both DHCP
backends, and each backend already had its own assertions on it. Neither
compared them, so the two could drift apart and both files would stay green.

The new block renders the same node through kea_build_node_reservations6 and
through the ISC omshell session and asserts the URL is the same string, that Kea
marks it always-send, and that a node asking for xnba gets none from either
backend.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 12:05:35 -03:00
Daniel Hilst a27f534d43 fix(packaging): an openSUSE node cannot fetch its loader over IPv6
xCAT.spec asked for the file /usr/sbin/in.tftpd. openSUSE has two providers and
the solver installed atftp, whose server has no IPv6 option: it binds
0.0.0.0:69 only, and --bind-address :: does not change that. A node that boots
over DHCPv6 therefore received its address and its RFC 5970 bootfile-url, asked
the server for the loader, timed out, released the address and started again.

xCAT.spec now requires the tftp package where suse_version is set. That is the
hpa server, and it listens on 0.0.0.0:69 and [::]:69 at once. EL keeps the file
dependency, which has one provider there, tftp-server.

Measured on a Leap 15.6 management node: with atftp the loader could not be
fetched over IPv6 and could over IPv4; after the swap both families served the
same 1372160 byte loader, and a missing file still answered File Not Found.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 12:03:56 -03:00
Daniel Hilst 139249f811 test(packaging): the openSUSE TFTP dependency can resolve to a server with no IPv6
xCAT.spec requires the file /usr/sbin/in.tftpd. Two openSUSE packages provide
it, and atftp has no IPv6 option at all.

The test asserts openSUSE asks for the tftp package by name and no longer leaves
a file dependency atftp can satisfy, and that EL keeps the file dependency it
resolves to tftp-server.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 12:03:56 -03:00
Daniel Hilst 231fadfaad fix(template): an IPv6 master is not bracketed in the autoyast media_url
The media_url of an autoyast add-on product came from
"#TABLE:noderes:$NODE:nfsserver#", which substitutes the address of the
management node without brackets. An IPv6-only node therefore received
<media_url>http://2001:db8::1//install/leap15.6/x86_64</media_url>, and YaST
cannot read that as a URL, so the installer reached no repository.

_sles_media_url() uses the bracketed master host for a node with no IPv4 address
and keeps the substitution for every other node, which service nodes rely on.

xCAT-test/unit/template_ipv6_rendering.t fails its two IPv6 assertions when the
bracketed host is dropped.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 11:07:37 -03:00
Daniel Hilst 706efff55b test(template): the autoyast media_url does not bracket an IPv6 master
Template.pm built the media_url of each add-on product from the noderes
substitution, which writes the address of the management node as it stands. An
IPv6-only node was told to install from "http://2001:db8::1//install/...".

The test covers the IPv6-only URL on the default port and on a custom port, and
holds down that a node with IPv4 keeps the substitution.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 11:07:37 -03:00
Daniel Hilst c7139d062c fix(grub2): an IPv6 facing address is substituted bare into the install URL
grub2.pm replaced !myipfn! with the facing address as it stands. sles.pm and
anaconda.pm put that placeholder where a URI authority expects a host, with the
port already appended, so an IPv6-only node was told to install from
"http://2001:db8:156::1:80/install/...". Nothing can split that into an address
and a port, so the installer could not reach the repository.

substitute_facing_address() brackets a placeholder that a port follows and
leaves every other one bare, because xcatd=!myipfn! names a host with no port
and the postscripts read that value as it stands.

The other five plugins that substitute the placeholder serve BIOS PXE, ppc64le
and yaboot, none of which network boots over IPv6, so they keep the bare
replacement.

grub2_ipv6.t and sles_install_url_ipv6.t fail their bracketing assertions when
the port-aware branch is removed, and their bare-address assertions still pass.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 10:46:39 -03:00
Daniel Hilst 19dc1de843 test(grub2): the facing address is substituted without brackets into a URL
grub2.pm replaces !myipfn! with the facing address of the management node. The
placeholder stands in front of ":$httpport" in the install URLs, so an IPv6
address was substituted bare and the authority could not be read.

grub2_ipv6.t now exercises setstate with a kcmdline that carries the placeholder
inside an install URL. sles_install_url_ipv6.t asserts the substitution itself
against the real module, in both contexts: bracketed where a port follows, bare
where none does.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 10:46:39 -03:00
Daniel Hilst a701dfc206 fix(sles): an IPv6 install server is not bracketed in the install URLs
sles.pm built its install URLs as "$server:$port". An IPv6 address contains
colons, so a node was told to install from
"http://2001:db8:156::1:80/install/leap15.6/x86_64/1" and could not reach the
repository. The same join was in the statelite and stateless imgurl=. The XCAT=
argument beside them was already bracketed.

rootimg_url() and autoyast_boot_args() now build those URLs through
xCAT::NetworkUtils::uri_authority(), which brackets an IPv6 address and passes a
name or a !myipfn! placeholder through unchanged. uri_authority is the former
xcatd_cmdline_address, renamed because three URL sites now share it;
xcatd_cmdline_address stays as a wrapper for its own callers.

xCAT-test/unit/sles_install_url_ipv6.t fails its four IPv6 assertions when the
colon join is restored.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 10:21:32 -03:00
Daniel Hilst c268c42475 test(sles): the install URLs do not bracket an IPv6 server
sles.pm joined the install server and the port with a colon, so an IPv6
management node produced "install=http://2001:db8:156::1:80/install/...". That
authority cannot be split into an address and a port.

The test covers the diskful autoyast= and install= arguments, the statelite and
stateless imgurl=, and the authority helper the three share.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 10:21:32 -03:00
Daniel Hilst c59021e408 fix(dhcp): no DHCPv6 configuration is written on openSUSE
ISC serves DHCPv4 and DHCPv6 from two files in the same directory. dhcp.pm set
the v6 file inside the branch that switches the v4 file to /etc/dhcp/dhcpd.conf.
openSUSE already has /etc/dhcpd.conf, so the branch never ran, $dhcp6conffile
stayed undef, and makedhcp wrote no IPv6 configuration. dhcpd6 then exited with
"No subnet6 declaration for eth1" and no node could get an IPv6 lease.

dhcp6_conf_file_for() now returns the v6 file that belongs beside a given v4
file, and the module derives it after every branch that chooses the v4 file.

xCAT-test/unit/dhcp_dhcpv6_conf_file.t fails on the openSUSE and /etc/dhcp3
layouts when the derivation is restricted to /etc/dhcp again.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 09:46:30 -03:00
Daniel Hilst 368026599d test(dhcp): makedhcp writes no DHCPv6 configuration on openSUSE
dhcp.pm derived the DHCPv6 configuration file only inside the branch that
switches the DHCPv4 file to /etc/dhcp/dhcpd.conf. openSUSE ships /etc/dhcpd.conf,
so that branch never ran and the v6 file stayed undef.

The test asserts the v6 file for each layout ISC uses, that an AIX file yields
none, and that the module derives the v6 file from the v4 file it chose.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 09:46:30 -03:00
Daniel Hilst cedafca382 fix(packaging): xcatd cannot start on openSUSE without Date::Parse
xcatd.pm uses Date::Parse. xCAT-server relies on the automatic dependency
generator for it, and EL does emit perl(Date::Parse). openSUSE has no
perl-generators, so nothing installs perl-TimeDate and every start of xcatd ends
with "Can't locate Date/Parse.pm in @INC at xCAT/xcatd.pm line 10". systemd
retries and the service never comes up.

xCAT-server.spec now requires the perl(Date::Parse) capability where
suse_version is set.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 09:20:25 -03:00
Daniel Hilst c5eed8e83f test(packaging): the openSUSE requires test covers only perl-xCAT
xcatd also fails to start on openSUSE, on Date::Parse, which xCAT-server ships in
xcatd.pm. The test named perl-xCAT alone, so that gap was not captured.

The expected capabilities are now a table keyed by spec, and xCAT-server is in
it. The set comes from comparing every perl() capability the EL build generates
against what a Leap 15.6 management node can resolve: of 88, only these five are
required and absent.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 09:20:25 -03:00
Daniel Hilst 3f171b2b15 fix(packaging): xcatconfig cannot find JSON.pm on openSUSE
perl-xCAT relies on the automatic Perl dependency generator for its module
requirements. openSUSE has no perl-generators, so the built rpm requires nothing
but /bin/sh and four modules xCAT loads are never installed: JSON, Expect, SNMP
and HTML::Form. Installing xCAT on an openSUSE management node then ends with
"Can't locate JSON.pm in @INC" from xCAT::Table, xcatconfig aborts, and xcatd
never starts.

perl-xCAT.spec now requires those four capabilities where suse_version is set.
EL is unchanged and keeps deriving them from the sources.

The four modules are the ones measured absent on a Leap 15.6 management node
that has every other xCAT dependency installed.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 08:58:59 -03:00
Daniel Hilst c559224455 test(packaging): perl-xCAT declares no Perl dependencies for openSUSE
perl-xCAT.spec carries no Requires and relies on the automatic Perl dependency
generator. EL has perl-generators and derives 99 perl() requires from the module
sources. openSUSE has none, so the built rpm requires only /bin/sh.

The test reads the Requires of perl-xCAT with rpmspec for openSUSE and for EL10.
On openSUSE it expects the four modules a Leap node does not otherwise carry. On
EL it expects none of them, because perl-generators derives them there.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 08:58:59 -03:00
Daniel Hilst d3e4d010db test(packaging): a bail-out in the openSUSE spec test stops the whole suite
opensuse_pod2man_buildrequires.t called BAIL_OUT when rpmspec exited non-zero.
prove stops every remaining file on a bail-out, not only the one that called it,
so one unreadable spec would hide the result of the rest of the suite.

The close failure now dies, which fails that file alone.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 08:35:40 -03:00
Daniel Hilst 7e12f8d538 fix(packaging): xCAT cannot install on openSUSE without perl-DB_File
xCAT-server requires perl-DB_File on every platform below EL10. openSUSE has no
package of that name -- it ships the module in perl-core-DB_File -- so zypper
answers "nothing provides 'perl-DB_File' needed by the to be installed
xCAT-server" and refuses to install xCAT on an openSUSE management node.

xCAT-server.spec now requires the perl(DB_File) capability where suse_version is
set, which perl-core-DB_File provides. EL9 keeps the package name EPEL
publishes, and EL10 keeps the weak requirement it has for riscv64.

xCAT-test/unit/opensuse_db_file_requires.t fails on the openSUSE assertions
without this change.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 08:35:29 -03:00
Daniel Hilst 7ec7c8ba03 test(packaging): xCAT-server asks openSUSE for a DB_File package it has not
openSUSE ships DB_File in perl-core-DB_File. The spec asks for perl-DB_File on
every platform below EL10, so zypper refuses the transaction with "nothing
provides 'perl-DB_File' needed by the to be installed xCAT-server" and the
management node cannot install xCAT.

The test reads the Requires of xCAT-server with rpmspec for openSUSE, EL9 and
EL10. On openSUSE it expects the perl(DB_File) capability and no perl-DB_File
package name. The EL assertions hold the existing behaviour down: EL9 keeps the
hard package requirement, EL10 keeps none.

Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
2026-10-28 08:35:22 -03:00