Merge branch 'main' into fix/setup-sigpipe-head

Author: osu

.github/actions/setup-mkosi-environment/action.yml

@@ -4,7 +4,7 @@ inputs:
   rust-version:
     description: 'Rust toolchain version to install'
     required: false
-    default: '1.90.0'
+    default: '1.96.0'
   arch:
     description: 'Target architecture (x86_64 or aarch64)'
     required: false

.github/workflows/build-boot-artifacts.yml

@@ -140,7 +140,7 @@ jobs:
         if: inputs.build_type == 'ephemeral'
         uses: ./.github/actions/setup-mkosi-environment
         with:
-          rust-version: '1.90.0'
+          rust-version: '1.96.0'
           arch: ${{ inputs.arch }}
 
       # Setup Docker for container-based builds

Cargo.lock

@@ -26,7 +26,7 @@ authors = ["NVIDIA Carbide Engineering <carbide-dev@exchange.nvidia.com>"]
 
 [workspace.dependencies]
 clap = { version = "4", features = ["derive", "env"] }
-libredfish = { git = "https://github.com/NVIDIA/libredfish.git", tag = "v0.44.10" }
+libredfish = { git = "https://github.com/NVIDIA/libredfish.git", tag = "v0.44.11" }
 librms = { git = "https://github.com/NVIDIA/nv-rms-client.git", tag = "v0.9.0-rc1" }
 ansi-to-html = "0.2.2"
 

Cargo.toml

@@ -46,7 +46,7 @@ BUILD_CONTAINER_AARCH64_URL = { value = "urm.nvidia.com/swngc-ngcc-docker-local/
 #
 # Make sure to update the RUST_NIGHTLY variable in
 # dev/docker/Dockerfile.build-container-x86_64 if you change this.
-RUST_NIGHTLY = "nightly-2026-05-27"
+RUST_NIGHTLY = "nightly-2026-06-16"
 
 [tasks.book]
 workspace = false
@@ -203,7 +203,7 @@ echo "Image carbide-build-x86_64 built. Run: cargo make cargo-docker -- build -p
 
 [tasks.build-cargo-docker-image-minimal]
 category = "Build"
-description = "Build minimal Cargo Docker image (Rust 1.90 + protoc). Quick (~2-5 min). Required once before cargo-docker-minimal for workspace builds (e.g. carbide-rpc)."
+description = "Build minimal Cargo Docker image (Rust 1.96 + protoc). Quick (~2-5 min). Required once before cargo-docker-minimal for workspace builds (e.g. carbide-rpc)."
 workspace = false
 script = '''
 set -e
@@ -213,7 +213,7 @@ echo "Image carbide-build-minimal built. Run: cargo make cargo-docker-minimal --
 
 [tasks.build-pxe-build-container]
 category = "Build"
-description = "Build the PXE artifact build container image (Ubuntu 24.04 + Rust 1.90 + mkosi deps). Required once before pxe-docker-x86."
+description = "Build the PXE artifact build container image (Ubuntu 24.04 + Rust 1.96 + mkosi deps). Required once before pxe-docker-x86."
 workspace = false
 script = '''
 set -e
@@ -301,7 +301,7 @@ fi
 
 [tasks.cargo-docker-minimal]
 category = "Build"
-description = "Run Cargo in minimal Linux image (Rust 1.90 + protoc). Requires build-cargo-docker-image-minimal once (~2-5 min). For DB tests set DATABASE_URL (use host.docker.internal, not localhost). Runs as root and chowns /code afterward so target/ stays writable on macOS/Colima."
+description = "Run Cargo in minimal Linux image (Rust 1.96 + protoc). Requires build-cargo-docker-image-minimal once (~2-5 min). For DB tests set DATABASE_URL (use host.docker.internal, not localhost). Runs as root and chowns /code afterward so target/ stays writable on macOS/Colima."
 workspace = false
 dependencies = ["check-cargo-home-set"]
 script = '''

Makefile.toml

@@ -26,8 +26,11 @@ use rpc::forge as forgerpc;
 #[clap(rename_all = "kebab-case")]
 pub enum NmxcOperationArg {
     ComputeNodeInfoList,
+    SwitchNodeInfoList,
     GpuInfo,
     GpuInfoList,
+    PartitionInfoList,
+    GetDomainProperties,
 }
 
 impl From<NmxcOperationArg> for forgerpc::NmxcBrowseOperation {
@@ -36,8 +39,15 @@ impl From<NmxcOperationArg> for forgerpc::NmxcBrowseOperation {
             NmxcOperationArg::ComputeNodeInfoList => {
                 forgerpc::NmxcBrowseOperation::ComputeNodeInfoList
             }
+            NmxcOperationArg::SwitchNodeInfoList => {
+                forgerpc::NmxcBrowseOperation::SwitchNodeInfoList
+            }
             NmxcOperationArg::GpuInfo => forgerpc::NmxcBrowseOperation::GpuInfo,
             NmxcOperationArg::GpuInfoList => forgerpc::NmxcBrowseOperation::GpuInfoList,
+            NmxcOperationArg::PartitionInfoList => forgerpc::NmxcBrowseOperation::PartitionInfoList,
+            NmxcOperationArg::GetDomainProperties => {
+                forgerpc::NmxcBrowseOperation::GetDomainProperties
+            }
         }
     }
 }
@@ -52,9 +62,18 @@ List the GPUs on a chassis via NMX-C:
 List the compute nodes on a chassis:
     $ nico-admin-cli browse nmxc --chassis-serial 1234567890 --operation compute-node-info-list
 
+List the switch nodes on a chassis:
+    $ nico-admin-cli browse nmxc --chassis-serial 1234567890 --operation switch-node-info-list
+
 Get info for a specific GPU UID:
     $ nico-admin-cli browse nmxc --chassis-serial 1234567890 --operation gpu-info --gpu-uid 42
 
+List NMX-C partitions:
+    $ nico-admin-cli browse nmxc --chassis-serial 1234567890 --operation partition-info-list
+
+Get NMX-C domain properties:
+    $ nico-admin-cli browse nmxc --chassis-serial 1234567890 --operation get-domain-properties
+
 ")]
 pub struct Args {
     #[clap(long, help = "Chassis serial number")]

crates/admin-cli/src/browse/nmxc/args.rs

@@ -21,7 +21,7 @@ pub mod proto {
 
 use carbide_rpc_utils::dhcp::{
     DhcpConfig as ModelDhcpConfig, HostConfig as ModelHostConfig,
-    InterfaceInfo as ModelInterfaceInfo,
+    InterfaceInfo as ModelInterfaceInfo, InterfaceInfoV6 as ModelInterfaceInfoV6,
 };
 use carbide_uuid::machine::MachineInterfaceId;
 use proto::dhcp_server_control_client::DhcpServerControlClient;
@@ -47,6 +47,28 @@ impl From<ModelDhcpConfig> for proto::DhcpConfig {
                 .collect(),
             carbide_provisioning_server_ipv4: c.carbide_provisioning_server_ipv4.to_string(),
             carbide_dhcp_server: c.carbide_dhcp_server.to_string(),
+            carbide_nameservers_v6: c
+                .carbide_nameservers_v6
+                .iter()
+                .map(|ip| ip.to_string())
+                .collect(),
+            carbide_ntpservers_v6: c
+                .carbide_ntpservers_v6
+                .iter()
+                .map(|ip| ip.to_string())
+                .collect(),
+            carbide_dhcp_server_v6: c.carbide_dhcp_server_v6.map(|ip| ip.to_string()),
+            dhcpv6_preferred_lifetime_secs: c.dhcpv6_preferred_lifetime_secs,
+            dhcpv6_valid_lifetime_secs: c.dhcpv6_valid_lifetime_secs,
+        }
+    }
+}
+
+impl From<ModelInterfaceInfoV6> for proto::InterfaceInfoV6 {
+    fn from(i: ModelInterfaceInfoV6) -> Self {
+        proto::InterfaceInfoV6 {
+            address: i.address.map(|ip| ip.to_string()),
+            prefix: i.prefix,
         }
     }
 }
@@ -60,6 +82,7 @@ impl From<ModelInterfaceInfo> for proto::InterfaceInfo {
             fqdn: i.fqdn,
             booturl: i.booturl,
             mtu: i.mtu,
+            ipv6: i.ipv6.map(Into::into),
         }
     }
 }

crates/agent/src/dhcp_server_grpc_client.rs

@@ -138,6 +138,52 @@ pub struct ServiceAddresses {
     pub nameservers: Vec<IpAddr>,
 }
 
+fn build_dhcp_ntp_servers(
+    nc: &rpc::ManagedHostNetworkConfigResponse,
+    service_addrs: &ServiceAddresses,
+) -> Vec<Ipv4Addr> {
+    // Start with the NTP servers from the service addresses, which is read from carbide-ntp.forge.
+    let mut ntp_servers = service_addrs
+        .ntpservers
+        .iter()
+        .filter_map(|x| match x {
+            IpAddr::V4(x) => Some(*x),
+            _ => None,
+        })
+        .collect::<Vec<Ipv4Addr>>();
+
+    // If the site has configured NTP servers, use them instead.
+    if !nc.ntp_servers.is_empty() {
+        let site_ntp_servers: Vec<Ipv4Addr> = nc.ntp_servers
+        .iter()
+        .filter_map(|s| match IpAddr::from_str(s) {
+            Ok(IpAddr::V4(ip)) => Some(ip),
+            Ok(IpAddr::V6(_)) => {
+                tracing::debug!(
+                    ntp_server = %s,
+                    "IPv6 NTP server from ManagedHostNetworkConfigResponse is ignored for DHCPv4 config"
+                );
+                None
+            }
+            Err(e) => {
+                tracing::debug!(
+                    ntp_server = %s,
+                    error = %e,
+                    "Invalid NTP server IP from ManagedHostNetworkConfigResponse, ignoring"
+                );
+                None
+            }
+        })
+        .collect();
+
+        if !site_ntp_servers.is_empty() {
+            ntp_servers = site_ntp_servers;
+        }
+    }
+
+    ntp_servers
+}
+
 /// How we tell HBN to notice the new file we wrote
 #[derive(Debug)]
 struct PostAction {
@@ -1029,14 +1075,7 @@ async fn update_dhcp_via_grpc(
         })
         .collect::<Vec<Ipv4Addr>>();
 
-    let ntpservers_v4 = service_addrs
-        .ntpservers
-        .iter()
-        .filter_map(|x| match x {
-            IpAddr::V4(x) => Some(*x),
-            _ => None,
-        })
-        .collect::<Vec<Ipv4Addr>>();
+    let ntpservers_v4 = build_dhcp_ntp_servers(network_config, service_addrs);
 
     let pxe_ip_v4 = service_addrs
         .pxe_ips
@@ -1445,14 +1484,7 @@ fn write_dhcp_v4_server_config(
         })
         .collect::<Vec<Ipv4Addr>>();
 
-    let ntpservers_v4 = service_addrs
-        .ntpservers
-        .iter()
-        .filter_map(|x| match x {
-            IpAddr::V4(x) => Some(*x),
-            _ => None,
-        })
-        .collect::<Vec<Ipv4Addr>>();
+    let ntpservers_v4 = build_dhcp_ntp_servers(nc, service_addrs);
 
     let pxe_ip_v4 = service_addrs
         .pxe_ips
@@ -1911,6 +1943,54 @@ mod tests {
         carbide_host_support::init_logging("nico-dpu-agent").unwrap();
     }
 
+    #[test]
+    fn test_build_dhcp_ntp_servers() {
+        let service_addrs = ServiceAddresses {
+            pxe_ips: vec![],
+            ntpservers: vec![IpAddr::from([192, 0, 2, 20])],
+            nameservers: vec![],
+        };
+        let nc = rpc::ManagedHostNetworkConfigResponse {
+            ntp_servers: vec!["198.51.100.1".to_string(), "198.51.100.2".to_string()],
+            ..Default::default()
+        };
+
+        let out = build_dhcp_ntp_servers(&nc, &service_addrs);
+        assert_eq!(
+            out,
+            vec![
+                Ipv4Addr::from([198, 51, 100, 1]),
+                Ipv4Addr::from([198, 51, 100, 2])
+            ]
+        );
+    }
+
+    #[test]
+    fn test_build_dhcp_ntp_servers_fallback() {
+        let service_addrs = ServiceAddresses {
+            pxe_ips: vec![],
+            ntpservers: vec![IpAddr::from([192, 0, 2, 20])],
+            nameservers: vec![],
+        };
+
+        let empty_nc = rpc::ManagedHostNetworkConfigResponse::default();
+
+        assert_eq!(
+            build_dhcp_ntp_servers(&empty_nc, &service_addrs),
+            vec![Ipv4Addr::from([192, 0, 2, 20])]
+        );
+
+        let invalid_nc = rpc::ManagedHostNetworkConfigResponse {
+            ntp_servers: vec!["not-an-ip".to_string(), "2001:db8::1".to_string()],
+            ..Default::default()
+        };
+
+        assert_eq!(
+            build_dhcp_ntp_servers(&invalid_nc, &service_addrs),
+            vec![Ipv4Addr::from([192, 0, 2, 20])]
+        );
+    }
+
     #[test]
     fn test_hostname() -> Result<(), Box<dyn std::error::Error>> {
         let syscall_h = super::hostname()?;
@@ -2935,6 +3015,7 @@ mod tests {
 
             // yes it's in there twice I dunno either
             dhcp_servers: vec!["10.217.5.197".to_string(), "10.217.5.197".to_string()],
+            ntp_servers: vec![],
             vni_device: "vxlan48".to_string(),
 
             managed_host_config: Some(netconf),
@@ -3382,6 +3463,7 @@ mod tests {
             rebinding_time_secs: 432000,
             carbide_api_url: None,
             carbide_dhcp_server: Ipv4Addr::from([10, 217, 5, 39]),
+            ..Default::default()
         };
 
         let mut network_config = rpc::ManagedHostNetworkConfigResponse {
@@ -3421,6 +3503,7 @@ mod tests {
 
             // yes it's in there twice I dunno either
             dhcp_servers: vec!["10.217.5.197".to_string(), "10.217.5.197".to_string()],
+            ntp_servers: vec![],
             vni_device: "vxlan48".to_string(),
 
             managed_host_config: Some(netconf),
@@ -3574,6 +3657,7 @@ mod tests {
             rebinding_time_secs: 432000,
             carbide_api_url: None,
             carbide_dhcp_server: Ipv4Addr::from([10, 217, 5, 39]),
+            ..Default::default()
         };
         let dhcp_contents = super::read_limited(g.path())?;
         assert!(dhcp_contents.contains("vlan196"));
@@ -3616,6 +3700,7 @@ mod tests {
             routing_profile: None,
             traffic_intercept_config: None,
             dhcp_servers: vec![],
+            ntp_servers: vec![],
             vni_device: "vxlan48".to_string(),
             managed_host_config: Some(netconf),
             managed_host_config_version: "V1-T1".to_string(),

crates/agent/src/ethernet_virtualization.rs

@@ -885,6 +885,7 @@ async fn handle_netconf(AxumState(state): AxumState<Arc<Mutex<State>>>) -> impl
         }),
 
         dhcp_servers: vec!["127.0.0.1".to_string()],
+        ntp_servers: vec![],
         vni_device: "".to_string(),
 
         managed_host_config: Some(rpc::forge::ManagedHostNetworkConfig {

crates/agent/src/tests/full.rs

@@ -20,6 +20,7 @@ applicable.
 | `ib_config` | `Option<IBFabricConfig>` | — | InfiniBand fabric configuration (see [IBFabricConfig](#ibfabricconfig)). |
 | `asn` | `u32` | **required** | Autonomous System Number, fixed per environment. Used by nico-dpu-agent for `frr.conf` BGP routing. |
 | `dhcp_servers` | `Vec<Ipv4Addr>` | `[]` | DHCP server addresses announced to DPUs during network provisioning. |
+| `ntp_servers` | `Vec<Ipv4Addr>` | `[]` | Site-level NTP server IPs used for BMC time configuration and DHCP NTP Server configuration. |
 | `route_servers` | `Vec<String>` | `[]` | Route server IPs for L2VPN Ethernet Virtual network support. |
 | `enable_route_servers` | `bool` | `false` | Enables route server injection into DPU FRR configs for L2VPN. |
 | `deny_prefixes` | `Vec<Ipv4Network>` | `[]` | IPv4 CIDR prefixes that tenant instances are blocked from reaching. Generates iptables DROP rules and nvue ACL policies. |
@@ -33,7 +34,7 @@ applicable.
 | `listen_mode` | `ListenMode` | `Tls` | Transport mode: `plaintext_http1`, `plaintext_http2`, or `tls`. |
 | `auth` | `Option<AuthConfig>` | — | Authentication/authorization settings (see [AuthConfig](#authconfig)). |
 | `pools` | `Option<HashMap<String, ResourcePoolDef>>` | — | Resource pools that allocate IPs, VNIs, etc. Required but `Option` for partial-config merging. |
-| `networks` | `Option<HashMap<String, NetworkDefinition>>` | — | Networks created at startup. Alternative: `CreateNetworkSegment` gRPC. |
+| `networks` | `Option<HashMap<String, NetworkDefinition>>` | — | Networks created at startup. Alternative: `CreateNetworkSegment` gRPC. `NetworkDefinition` supports dual-stack seed-time segments with optional `prefix_v6` and `dhcpv6_link_address`; config edits do not retrofit prefixes onto an already-seeded segment because seed definitions are snapshotted on first create. |
 | `dpu_ipmi_tool_impl` | `Option<String>` | — | IPMI tool implementation for DPU power control (`"prod"` or `"fake"`). |
 | `dpu_ipmi_reboot_attempts` | `Option<u32>` | — | Retry count when IPMI errors during DPU reboot. |
 | `bmc_session_lockout_threshold` | `u32` | `3` | Consecutive BMC HTTP 401/403 responses before session-token login attempts stop for that BMC. |