diff --git a/src/libnm-platform/nm-platform.h b/src/libnm-platform/nm-platform.h index b19b933f2a..a4cba9ced8 100644 --- a/src/libnm-platform/nm-platform.h +++ b/src/libnm-platform/nm-platform.h @@ -430,60 +430,6 @@ typedef union { #define __NMPlatformIPRoute_COMMON \ __NMPlatformObjWithIfindex_COMMON; \ \ - /* The NMIPConfigSource. For routes that we receive from cache this corresponds - * to the rtm_protocol field (and is one of the NM_IP_CONFIG_SOURCE_RTPROT_* values). - * When adding a route, the source will be coerced to the protocol using - * nmp_utils_ip_config_source_coerce_to_rtprot(). - * - * rtm_protocol is part of the primary key of an IPv4 route (meaning, you can add - * two IPv4 routes that only differ in their rtm_protocol. For IPv6, that is not - * the case. - * - * When deleting an IPv4/IPv6 route, the rtm_protocol field must match (even - * if it is not part of the primary key for IPv6) -- unless rtm_protocol is set - * to zero, in which case the first matching route (with proto ignored) is deleted. */ \ - NMIPConfigSource rt_source; \ - \ - guint8 plen; \ - \ - /* RTA_METRICS: - * - * For IPv4 routes, these properties are part of their - * ID (meaning: you can add otherwise identical IPv4 routes that - * only differ by the metric property). - * On the other hand, for IPv6 you cannot add two IPv6 routes that only differ - * by an RTA_METRICS property. - * - * When deleting a route, kernel seems to ignore the RTA_METRICS properties. - * That is a problem/bug for IPv4 because you cannot explicitly select which - * route to delete. Kernel just picks the first. See rh#1475642. */ \ - \ - /* RTA_METRICS.RTAX_LOCK (iproute2: "lock" arguments) */ \ - bool lock_window : 1; \ - bool lock_cwnd : 1; \ - bool lock_initcwnd : 1; \ - bool lock_initrwnd : 1; \ - bool lock_mtu : 1; \ - \ - /* if TRUE, the "metric" field is interpreted as an offset that is added to a default - * metric. For example, form a DHCP lease we don't know the actually used metric, because - * that is determined by upper layers (the configuration). However, we have a default - * metric that should be used. So we set "metric_any" to %TRUE, which means to use - * the default metric. However, we still treat the "metric" field as an offset that - * will be added to the default metric. In most case, you want that "metric" is zero - * when setting "metric_any". */ \ - bool metric_any : 1; \ - \ - /* like "metric_any", the table is determined by other layers of the code. - * This field overrides "table_coerced" field. If "table_any" is true, then - * the "table_coerced" field is ignored (unlike for the metric). */ \ - bool table_any : 1; \ - /* Meta flags not honored by NMPlatform (netlink code). Instead, they can be - * used by the upper layers which use NMPlatformIPRoute to track routes that - * should be configured. */ \ - /* Whether the route should be committed even if it was removed externally. */ \ - bool r_force_commit : 1; \ - \ /* rtnh_flags * * Routes with rtm_flags RTM_F_CLONED are hidden by platform and @@ -525,16 +471,69 @@ typedef union { * zero (RT_TABLE_UNSPEC) are swapped, so that the default is the main * table. Use nm_platform_route_table_coerce()/nm_platform_route_table_uncoerce(). */ \ guint32 table_coerced; \ + /* The NMIPConfigSource. For routes that we receive from cache this corresponds + * to the rtm_protocol field (and is one of the NM_IP_CONFIG_SOURCE_RTPROT_* values). + * When adding a route, the source will be coerced to the protocol using + * nmp_utils_ip_config_source_coerce_to_rtprot(). + * + * rtm_protocol is part of the primary key of an IPv4 route (meaning, you can add + * two IPv4 routes that only differ in their rtm_protocol. For IPv6, that is not + * the case. + * + * When deleting an IPv4/IPv6 route, the rtm_protocol field must match (even + * if it is not part of the primary key for IPv6) -- unless rtm_protocol is set + * to zero, in which case the first matching route (with proto ignored) is deleted. */ \ + NMIPConfigSource rt_source; \ + \ + /* RTA_METRICS: + * + * For IPv4 routes, these properties are part of their + * ID (meaning: you can add otherwise identical IPv4 routes that + * only differ by the metric property). + * On the other hand, for IPv6 you cannot add two IPv6 routes that only differ + * by an RTA_METRICS property. + * + * When deleting a route, kernel seems to ignore the RTA_METRICS properties. + * That is a problem/bug for IPv4 because you cannot explicitly select which + * route to delete. Kernel just picks the first. See rh#1475642. */ \ + \ + /* RTA_METRICS.RTAX_LOCK (iproute2: "lock" arguments) */ \ + bool lock_window : 1; \ + bool lock_cwnd : 1; \ + bool lock_initcwnd : 1; \ + bool lock_initrwnd : 1; \ + bool lock_mtu : 1; \ + \ + /* if TRUE, the "metric" field is interpreted as an offset that is added to a default + * metric. For example, form a DHCP lease we don't know the actually used metric, because + * that is determined by upper layers (the configuration). However, we have a default + * metric that should be used. So we set "metric_any" to %TRUE, which means to use + * the default metric. However, we still treat the "metric" field as an offset that + * will be added to the default metric. In most case, you want that "metric" is zero + * when setting "metric_any". */ \ + bool metric_any : 1; \ + \ + /* like "metric_any", the table is determined by other layers of the code. + * This field overrides "table_coerced" field. If "table_any" is true, then + * the "table_coerced" field is ignored (unlike for the metric). */ \ + bool table_any : 1; \ + /* Meta flags not honored by NMPlatform (netlink code). Instead, they can be + * used by the upper layers which use NMPlatformIPRoute to track routes that + * should be configured. */ \ + /* Whether the route should be committed even if it was removed externally. */ \ + bool r_force_commit : 1; \ \ /* rtm_type. * * This is not the original type, if type_coerced is 0 then * it means RTN_UNSPEC otherwise the type value is preserved. */ \ + guint8 type_coerced; \ + \ /* Don't have a bitfield as last field in __NMPlatformIPAddress_COMMON. It would then * be unclear how the following fields get merged. We could also use a zero bitfield, * but instead we just have there the uint8 field. */ \ - guint8 type_coerced; \ + guint8 plen; \ ; typedef struct {