For new uses of nm_uuid_generate_from_strings() we should generate version5
UUIDs and we should use unique namespace UUID arguments.
The namespace UUID was so far replaced by always passing a special prefix
as first string. It seems nicer to use a namespace instead.
Version3 UUIDs should not be used for new applications.
Hence, nm_uuid_generate_from_strings_v3() is no longer a desirable way to
generate UUIDs, so drop the wrapper.
nm_uuid_generate_from_strings() uses variant3 UUIDs based on MD5.
We shouldn't use that in the future.
We will add a replacement, so rename this function so that the "good"
name is free again. Of course, code that uses this function currently
relies on that the behavior doesn't change. We cannot just drop it
entirely, but will replace it by something that gives the same result.
Rename.
If we're generating a connection for an externally configured slave,
refer the master by the UUID instead of the device name.
This doesn't matter most of the time. However, on a checkpoint restore
we need to make sure that a connection that is unambiguously the original
master is up.
Otherwise it could happen that a different connection was activated on the
same master device and the slaves being restored don't agree on which master
connection to bring up.
I can't think of any thing that would rely on this but I've been wrong
about more serious things before.
Fixes-test: @libnm_snapshot_reattach_unmanaged_ports_to_bridge
https://bugzilla.redhat.com/show_bug.cgi?id=2125615
If we're notified of a master appearing, make sure there's actually an
ifindex we're waiting for.
Triger an assertion failure if that is not the case, cause that's pretty
messed up.
Since commit a1de6810df ('device: don't ignore external slave removals')
we don't leave device_recheck_slave_status() on un-eslaving (that is
plink->master = 0) early enough.
This results in hooking of NM_MANAGER_DEVICE_IFINDEX_CHANGED even
when we're not actually waiting for any master device to come up,
accompanied by a messed up log entry:
device[3fa7cfc200be4e84] (portXc): enslaved to unknown device 0 (??)
We also log nonsense when we see any device's link being removed:
device[a9a4b65bde851bcf] (br0): ifindex: set ifindex 0 (old-l3cfg: 05c6a4409f84d9d2)
device[45d34e95fb71cce0] (portXa): master br0 with ifindex 0 appeared
We don't do further damage afterwards, so this is purely a cosmetic
annoyance.
This is something that does happen.
Is that a bug? If so, this should not be a warning message but an
assertion failure. If it's not a bug, then this does not warrant warning
level, because the user wouldn't know what to do about this and it's
something that occasionally happens.
Granted, the state handling in NMDevice is complex, that it's unclear
whether this indicates a problem or not. In any case, having a warning
does only confuse the user.
Have "len" before "elem_size". That is consistent with g_qsort_with_data()
and bsearch(), and is also what I would expect.
Note that the previous commit just renamed the function. If a user
of the new, changed API gets backported to an older branch, we will
get a compilation error and note that the arguments need to be adjusted.
The "nm_utils_" prefix is just too verbose. Drop it.
Also, Posix has a bsearch function. As this function
is similar, rename it.
Note that currently the arguments are provided in differnt
order from bsearch(). That will be partly addressed next.
That is the main reason for the rename. The next commit
will swap the arguments, so do a rename first to get a compilation
error when backporting a patch that uses the changed API.
find_parent_device_for_connection() must return a parent device even
if it's unrealized. In fact, callers already handle the unrealized
case; in specific:
- _internal_activate_device() will try to autoactivate a connection
on the unrealized parent to realize it;
- system_create_virtual_device()'s goal is to create a NMDevice
object, so it doesn't matter whether the parent is realized or not.
Relax the condition about managed-ness, since any unrealized
device is also unmanaged.
This change fixes the following scenario, where all profiles have
autoconnect=no and autoconnect-slaves=yes:
vrf0
-------------^----------------
| | |
| bridge0 bond0.4000
| .
bond0 <......................
---^---
| |
veth0 veth1
----> = controller
....> = VLAN parent
When profiles are added, unrealized devices are created for bond0 and
bridge0, but not for bond0.4000 becase its parent is unrealized. Then
the autoconnect-slaves machinery for vrf0 tries to activate all ports
but fails for bond0.4000 because it can't find a device for it.
https://bugzilla.redhat.com/show_bug.cgi?id=2101317
When disposing NMPolicy all the devices in the devices hash-table should
be unregistered and removed from the hash-table.
Fixes: 7e3d090acb ('policy: refactor tracking of registered devices')
We already have src/linux-headers, where we have complete copies of linux
user space headers. Of course that exists, because we want to use certain
features and don't depend on the installed kernel headers. Which works
well, because kernel user space API is stable, and we anyway want to
support compiling against a newer kernel and run against an older (e.g.
in a container). So having our copy of newer kernel headers is merely
as if we compiled against as newer kernel.
Add "src/nm-compat-headers" which has a similar purpose, but a different
approach. Instead of replacing the included header entirely, include
the system header and patch it with #define.
Use this for "linux/if_addr.h". Of course, the approach here is that we
no longer include <linux/if_addr.h> directly, but instead include
"nm-compat-headers/linux/if_addr.h".
Sync/blocking methods are ugly. Their name should highlight this.
Also, we may have an async variant, so we will need the "good" name
for apply() and apply_finish().
Previously, we used _nm_utils_ascii_str_to_bool(). That can accept any
kind of input (like "true"), so one might think that this is better to
use on user-input. However, NMSettingBond already validates the these
options are integers (either "0" or "1"). So a value like "true"
could never be here.
Use _nm_setting_bond_opt_value_as_intbool() because that asserts that
the option if of the expected type (integer).
These variants provide additional nm_assert() checks, and are thus
preferable.
Note that we cannot just blindly replace &g_array_index() with
&nm_g_array_index(), because the latter would not allow getting a
pointer at index [arr->len]. That might be a valid (though uncommon)
usecase. The correct replacement of &g_array_index() is thus
nm_g_array_index_p().
I checked the code manually and replaced uses of nm_g_array_index_p()
with &nm_g_array_index(), if that was a safe thing to do. The latter
seems preferable, because it is familar to &g_array_index().
Add nm_g_array_index() as a replacement for g_array_index(). The value
of nm_g_array_index(), nm_g_array_index_p(), nm_g_array_first() and
nm_g_array_last() is that they add nm_assert() checks for valid
parameters.
nm_g_array_{first,last}() now returns an lvalue and not a pointer.
As such, they are just shorthands for nm_g_array_index() at index
0 and len-1, respectively.
`nm_g_array_index_p(arr, Type, idx)` is almost the same as
`&nm_g_array_index(arr, Type, idx)`. The only difference (and why the
former variant exists), is that nm_g_array_index_p() allows to get a
pointer one after the end.
This means, this is correct and valid to do:
// arr->len might be zero
arr = nm_g_array_index_p(arr, Type, 0);
for (i = 0; i < arr->len; i++, arr++)
...
ptr = nm_g_array_index_p(arr, Type, 0);
end = nm_g_array_index_p(arr, Type, arr->len);
for (; ptr < end; ptr++)
...
This would not be valid to do with nm_g_array_{index,first,last}().
Also fix supporting "const GArray *arr" parameter. Of course, the function
casts the constness away. Technically, that matches the fact that arr->data
is also not a const pointer. In practice, we might want to propagate the
constness of the container to the constness of the element lookup. While
doable, that is not implemented.
Bond option netlink support requires primary property to be a ifindex
instead of the interface name. This is a workaround for supporting
specifying a primary that does not exist yet.
```
nmcli con add type bond ifname mybond0 bond.options "mode=active-backup,primary=veth1"
Connection 'bond-mybond0' (38100ef9-11e2-4003-aff9-cb2d152ce34f) successfully added.
nmcli con add type ethernet ifname veth1 master mybond0
cat /sys/class/net/mybond0/bonding/primary
veth1
```
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/1362
Fixes: e064eb9d13 ('bond: use netlink to set bond options')
_nm_config_data_log_sort() is used for sorting the groups in the
keyfile during nm_config_data_log(). The idea is to present the keyfile
in a defined, but useful order.
However, it is not a total order. That is, it will return c=0 (equal) for
certain groups, if the pre-existing order in the GKeyFile should be
honored. For example, we want to sort all [device*] sections close to
each other, but we want to preserve their relative order. In that case,
the function would return 0 although the group names differed.
Also, _nm_config_data_log_sort() does not expect to receive duplicate names.
It would return c!=0 for comparing "device" and "device".
This means, _nm_config_data_log_sort() is fine for sorting the input as
we have it. However, it cannot be used to binary search the groups. This
caused that some sections might be duplicated in the `NetworkManager
--print-config` output. Otherwise, it had no bad effects.
Fixes(no-backport): 78d34d7c2e ('config: fix printing default values for missing sections')
Previously, nm_platform_ip_address_sync() would always add the "IFA_F_NOPREFIXROUTE"
flag. Add a way to let the caller control that.
Add a flags argument, with a new flag "with-noprefixroute". By default
(with flags "none"), nm_platform_ip_address_sync() would no longer
add "IFA_F_NOPREFIXROUTE" flag, but the caller can now opt-in to that.
The purpose is that on "lo" interface we will want to let kernel
handle the prefix route. So have a per-ifindex opt-in for controlling
this.
During nm_platform_ip_address_flush() we use "none" flags, because the
function anyway doesn't add any addresses, so it wouldn't matter.
There is no change in behavior.
Co-authored-by: Thomas Haller <thaller@redhat.com>
We've been outright ignoring master-slave checks if the link ended up
without a master since commit 2e22880894 ('device: don't remove the
device from master if its link has no master').
This was done to deal with OpenVSwitch port-interface relationship,
where the interface's platform link lacked an actual master in platform
(what matters there is the OVSDB entry), but the fix was too wide.
Let's limit the special case to devices whose were not enslaved to
masters that lack a platform link, which pretty much happens for
OpenVSwitch only.
Morale: Write better commit messages of future you is going to be upset
Fixes: 2e22880894 ('device: don't remove the device from master if its link has no master')
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/1358
For the ipoib connection, it is still considered as valid if the
profile does not set the device name. Also, the ifcfg reader should not
duplicate the checks that `nm_connection_verify()` performs (especially
not wrongly). Therefore, NM should skip validating the DEVICE when
reading the ifcfg file for the ipoib connection.
https://bugzilla.redhat.com/show_bug.cgi?id=2122703
When writing the p-key setting to the ifcfg file and reading the
setting back, the value has to be consistent. This is not limited to
p-key only, any setting value during the ifcfg write and read also has
to be consistent.
This was probably added in commit cb5606cf1c ('ifcfg-rh:
add support for Infiniband partitions') as this is also what
ifup-ib does ([1]). For NetworkManager profiles however, the
p-key is also valid without the high bit set, so the ifcfg-rh
reader must honor that.
[1] 0c9fb6ca7b/rdma.ifup-ib (L75)
