For historic reasons is NMSettingBond implemented differently from other
settings. It uses a strdict, and adds some validation on top of that.
The idea was probably to be able to treat bond options more generically.
But in practice we cannot treat them as opaque values, but need to know,
validate and understand all the options. Thus, this implementation with a
strdict is not nice.
The user can set the GObject property NM_SETTING_BOND_OPTIONS to any
strdict, and the setter performs no validation or normalization. That
is probably good, because g_object_set() cannot return an error to
signalize invalid settings. As often, we have corresponding C API like
nm_setting_bond_add_option() and nm_setting_bond_remove_option(). It
should be possible to get the same result both with the C API and with
the GObject property setting. Since there is already a way to set
certain invalid values, it does not help if the C API tries to prevent
that. That implies, that also add-option does not perform additional
validation and sets whatever the user asks.
Remove all validation from nm_setting_bond_add_option() and
nm_setting_bond_remove_option(). This validation was anyway only very
basic. It was calling nm_setting_bond_validate_option(), which can check
whether the string is (for example) and integer, but it cannot do
validation beyond one option. In most cases, the validation needs to
take into account the bond mode or other options, so validating one
option in isolation is not very useful.
Proper validation should instead be done via nm_connection_verify().
However, due to another historic oddity, that verification is very
forgiving too and doesn't reject many invalid settings when it should.
That is hard to fix, because making validation more strict can break
existing (and working) configurations. However, verify() already contains
basic validation via nm_setting_bond_validate_option(). So in the previous
behavior nm_setting_bond_add_option() would silently do nothing (only
returning %FALSE) for invalid options, while now it would add the
invalid options to the dictionary -- only to have it later fail validation
during nm_connection_verify(). That is a slight change in behavior, however it
seems preferable.
It seems preferable and acceptable because most users that call
nm_setting_bond_add_option() already understand the meaning and valid
values. Keyfile and ifcfg-rh readers are the few exceptions, which really just
parse a string dictionary, without need to understand them. But nmtui
or nmstate already know the option they want to set. They don't expect
a failure there, nor do they need the validation.
Note that this change in behavior could be dangerous for example for the
keyfile/ifcfg-rh readers, which silently ignored errors before. We
don't want them to start failing if they read invalid options from a
file, so instead let those callers explicitly pre-validate the value
and log an warning.
https://bugzilla.redhat.com/show_bug.cgi?id=1887523
Change the default DNS priority of VPNs to -50, to avoid leaking
queries out of full-tunnel VPNs.
This is a change in behavior. In particular:
- when using dns=default (i.e. no split-dns) before this patch both
VPN and the local name server were added (in this order) to
resolv.conf; the result was that depending on resolv.conf options
and resolver implementation, the name servers were tried in a
certain manner which does not prevent DNS leaks.
With this change, only the VPN name server is added to resolv.conf.
- When using a split-dns plugin (systemd-resolved or dnsmasq), before
this patch the full-tunnel VPN would get all queries except those
ending in a local domain, that would instead be directed to the
local server.
After this patch, the VPN gets all queries.
To revert to the old behavior, set the DNS priority to 50 in the
connection profile.
Previously, both nm_setting_connection_add_permission() and the GObject
property setter would merely assert that the provided values are valid
(and otherwise don't do anything). That is bad for handling errors.
For example, we use the property setter to initialize the setting from
keyfile and GVariant (D-Bus). That means, if a user provides an invalid
permissions value, we would emit a g_critical() assertion failure, but
otherwise ignore the configuration. What we instead need to do is to
accept the value, and afterwards fail verification. That way, a proper error
message can be generated.
$ mcli connection add type ethernet autoconnect no ifname bogus con-name x connection.permissions 'bogus:'
(process:429514): libnm-CRITICAL **: 12:12:00.359: permission_new: assertion 'strchr (uname, ':') == NULL' failed
(process:429514): libnm-CRITICAL **: 12:12:00.359: nm_setting_connection_add_permission: assertion 'p != NULL' failed
Connection 'x' (2802d117-f84e-44d9-925b-bfe26fd85da1) successfully added.
$ $ nmcli -f connection.permissions connection show x
connection.permissions: --
While at it, also don't track the permissions in a GSList. Tracking one
permission in a GSList requires 3 allocations (one for the user string,
one for the Permission struct, and one for the GSList struct). Instead,
use a GArray. That is still not great, because GArray cannot be embedded
inside NMSettingConnectionPrivate, so tracking one permission also
requires 3 allocations (which is really a fault of GArray). So, GArray
is not better in the common case where there is only one permissions. But even
in the worst case (only one entry), GArray is no worse than GSList.
Also change the API of nm_setting_connection_add_permission().
Previously, the function would assert that the arguments are in
a certain form (strcmp (ptype, "user") == 0), but still document
the such behaviors like regular operation ("[returns] %FALSE if @ptype
or @pitem was invalid"). Don't assert against the function arguments.
Also, if you first set the user to "fo:o", then
nm_setting_connection_add_permission() would accept it -- only at
a later phase, the property setter would assert against such values.
Also, the function would return %FALSE both if the input value was
invalid (an error) and if the value already existed. I think the
function should not treat a duplicate entry like a badly formatted
input.
Now the function does much less asserting of the arguments, but will
return %FALSE only if the values are invalid. And it will silently ignore
duplicate entries.
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/636
Run:
./contrib/scripts/nm-code-format.sh -i
./contrib/scripts/nm-code-format.sh -i
Yes, it needs to run twice because the first run doesn't yet produce the
final result.
Signed-off-by: Antonio Cardace <acardace@redhat.com>
"active_slave" is by now deprecated and became an alias for "primary".
If a profile specifies both properties, only "primary" is honored, without
failing validation (to not break existing behavior).
Maybe we should introduce a normalization for such cases. But normalize
might not do the right thing, if a profile currently has "primary" set,
and the user modifies it to set "active_slave" to a different value,
normalize would not know which setting was set first and remove
"active_slave" again.
In the past, nm_setting_bond_add_option() performed some simple
normalization, but this was dropped, because (such incompatible) settings
can also be created via the GObject property. Our C accessor function
should not be less flexible than other ways of creating a profile.
In the end, whenever a user (or a tool) creates a profile, the tool must
be aware of the semantics. E.g. setting an IP route without a suitable
IP address is unlike to make sense, the tool must understand what it's
doing. The same is true for the bond options. When a tool (or user) sets
the "active_slave" property, then it must clear out the redundant
information from the "primary" setting. There is no alternative to this
problem than having tools smart enough to understand what they are
doing.
WireGuard's wg-quick primarily wants to avoid DNS leaks, and thus also
our import code should generate profiles that configure exclusive DNS
servers. This is done by setting "ipv[46].dns-priority" to a negative
value.
Note that if a profile leaves the DNS priority at zero (which in many
regard is the default), then the zero translates to 50 (for VPN
profiles) and 100 (for other profiles).
Instead of setting the DNS priority to -10, set it to -50. This gives
some more room so that the user can choose priorities that are worse
than the WireGuard's one, but still negative (exclusive). Also, since
the positive range defaults to 50 and 100, let's stretch the range a
bit.
Since this only affects import and creation of new profiles, such a
change in behavior seems acceptable.
A profile can configure "connection.wait-device-timeout" to indicate
that startup complete is blocked until a suitable device around.
This is useful for NetworkManager-wait-online and initrd mode.
Previously, we looked at NMPlatform whether a link with matching
interface-name was present. That is wrong because it cannot handle
profiles that rely on "ethernet.mac-address" setting or other "match"
settings. Also, the mere presence of the link does not yet mean
that the NMDevice was created and ready. In fact, there is a race here:
NMPlatform indicates that the device is ready (unblocking NMSettings),
but there is no corresponding NMDevice yet which keeps NetworkManager
busy to block startup complete.
Rework this. Now, only check whether there is a compatible device for
the profile.
Since we wait for compatible devices, it works now not only for the
interface name. Note that we do some optimizations so that we don't have
to re-evaluate all profiles (w.r.t. all devices) whenever something on the
device changes: we only care about this when all devices finally become
ready.
Also, we no longer start the timeout for "connection.wait-device-timeout"
when the profile appears. Instead, there is one system-wide start time
(NMSettingsPrivate.startup_complete_start_timestamp_msec). That simplifies
code and makes sense: we start waiting when NetworkManager is starting, not
when the profile gets added. Also, we wait for all profiles to become
ready together.
Before 1.24, nm_setting_bond_add_option() would clear
miimon/arp_interval settings when the respective other was set.
That was no longer done, with the effect that enabling (for example)
miimon on a bond profile that has arp_interval enabled, sets both
conflicting options.
That is not a severe problem, because the profile still validates.
However, at runtime only one of the settings can be actually configured.
Fix that, by restoring the previous behavior for the client. But note
that this time it's implemented in the client, and not in libnm's
nm_setting_bond_add_option().
Kernel (sysfs) and iproute2 only use numbers for the multicast_router
option. It's confusing that we name the options differently. Anyway,
that cannot be changed anymore. Clarify the meanings in the
documentation.
https://bugzilla.redhat.com/show_bug.cgi?id=1845608
For simple matches like match.interface-name, match.driver, and
match.path, arguably what we had was fine. There each element
(like "eth*") is a wildcard for a single name (like "eth1").
However, for match.kernel-command-line, the elements match individual
command line options, so we should have more flexibility of whether
a parameter is optional or mandatory. Extend the syntax for that.
- the elements can now be prefixed by either '|' or '&'. This makes
optional or mandatory elements, respectively. The entire match
evaluates to true if all mandatory elements match (if any) and
at least one of the optional elements (if any).
As before, if neither '|' nor '&' is specified, then the element
is optional (that means, "foo" is the same as "|foo").
- the exclamation mark is still used to invert the match. If used
alone (like "!foo") it is a shortcut for defining a mandatory match
("&!foo").
- the backslash can now be used to escape the special characters
above. Basically, the special characters ('|', '&', '!') are
stripped from the start of the element. If what is left afterwards
is a backslash, it also gets stripped and the remainder is the
pattern. For example, "\\&foo" has the pattern "&foo" where
'&' is no longer treated specially. This special handling of
the backslash is only done at the beginning of the element (after
the optional special characters). The remaining string is part
of the pattern, where backslashes might have their own meaning.
This change is mostly backward compatible, except for existing matches
that started with one of the special characters '|', '&', '!', and '\\'.
Add a new "path" property to the match setting, which can be used to
restrict a connection to devices with a given hardware path. The new
property is a list of patterns that are matched against the ID_PATH
udev property of devices.
ID_PATH represents the topological persistent path of a device and
typically contains a subsystem string (pci, usb, platform, etc.) and a
subsystem-specific identifier. Some examples of paths are:
pci-0000:00:02.0
pci-0000:00:14.0-usb-0:5:1.0
platform-1c40000.ethernet
systemd-networkd also has a "Path=" option to match a device by udev
ID_PATH.
This does not match kernel behavior. You seem to be able to configure
multicast_snooping=auto,enabled with multicast_snooping off just fine.
Possibly this constrained was inspired by `ip link`, which says:
mcast_router MULTICAST_ROUTER - set bridge's multicast router if IGMP
snooping is enabled.
But kernel doesn't enforce this:
ip link delete br0 2>/dev/null; \
ip link add br0 type bridge mcast_router 1 mcast_snooping 0; \
grep ^ /sys/devices/virtual/net/br0/bridge/{multicast_router,multicast_snooping}
gives:
/sys/devices/virtual/net/br0/bridge/multicast_router:1
/sys/devices/virtual/net/br0/bridge/multicast_snooping:0
We probably should not implement additional constrains on top of what
kernel does, if the conditions are that obscure.
Fixes: e01d3b4c2b ('nm-setting-bridge: add 'multicast-router' bridge option')
"nm-settings-docs-nmcli.xml" will be generated by a tool that depends on
"clients/common/". The file should thus not be in libnm directory, otherwise
there is a circular dependency.
Move the file to "man/" directory.
For consistency, also move "nm-settings-docs-dbus.xml". Note that we
cannot move "nm-settings-docs-gir.xml" to "man/", because that one is
needed for building clients.
The name is bad. For one, we will have more files of the same format
("nm-settings-docs-nmcli.xml").
Also, "libnm/nm-settings-docs.xml" and "libnm/nm-property-docs.xml" had
basically the same file format. Their name should be similar.
Also the tool to generate the file should have a name that reminds to
the file that it creates.
The 'peer' property of ovs-patch is inserted into the 'options' column
of the ovsdb 'Interface' table. The ovs-vswitchd.conf.db man page says
about it:
options : peer: optional string
The name of the Interface for the other side of the patch. The
named Interface’s own peer option must specify this Interface’s
name. That is, the two patch interfaces must have reversed name
and peer values.
Therefore, it is wrong to validate the peer property as an IP address
and document it as such.
Fixes: d4a7fe4679 ('libnm-core: add ovs-patch setting')
Reported by coverity:
>>> CID 210213 Uninitialized pointer read (UNINIT)
>>> Using uninitialized value iter when calling
_nm_auto_free_variant_iter
Fixes: df1d214b2e ('clients: polkit-agent: implement polkit agent without using libpolkit')
Reported by coverity:
>>> CID 210217: (UNINIT)
>>> Using uninitialized value "identities_gvariant" when calling
"gs_local_variant_unref".
Fixes: df1d214b2e ('clients: polkit-agent: implement polkit agent without using libpolkit')
Add a new "driver" match option to nm-settings. It allows to disable a
network connection configuration if a pattern is found or is not found
in the device driver name.
Add a new "kernel-command-line" match option to nm-settings. It allows
to disable a network connection configuration if a pattern is found or
is not found in /proc/cmdline.
