libkrun, vsock: Add explicit vsock configuration API

Refactor the vsock device to explicitly configure usage of INET and UNIX
proxies.

Add new public API functions for explicit vsock control:
- krun_disable_implicit_vsock(): Disable the implicit vsock device
- krun_add_vsock(): Add vsock with explicit TSI feature flags

krun_add_vsock() requires krun_disable_implicit_vsock() to be called
first, otherwise an error is returned - we only support 1 vsock device.

Add a check in krun_set_port_map() and krun_add_vsock_port() to ensure vsock is
enabled.

Signed-off-by: Matej Hrica <[email protected]>

Author: mtjhrc

include/libkrun.h

@@ -273,6 +273,10 @@ int32_t krun_add_virtiofs2(uint32_t ctx_id,
    as required by gvproxy in vfkit mode. */
 #define NET_FLAG_VFKIT 1 << 0
 
+/* TSI (Transparent Socket Impersonation) feature flags for vsock */
+#define KRUN_TSI_HIJACK_INET  (1 << 0)
+#define KRUN_TSI_HIJACK_UNIX  (1 << 1)
+
 /* Taken from uapi/linux/virtio_net.h */
 #define NET_FEATURE_CSUM 1 << 0
 #define NET_FEATURE_GUEST_CSUM 1 << 1
@@ -808,6 +812,27 @@ int32_t krun_add_vsock_port2(uint32_t ctx_id,
                              uint32_t port,
                              const char *c_filepath,
                              bool listen);
+
+/**
+ * Add a vsock device with specified TSI features.
+ *
+ * By default, libkrun creates a vsock device implicitly with TSI hijacking
+ * enabled based on heuristics. Calling this function overrides the implicit
+ * behavior and explicitly configures the vsock device.
+ *
+ * Currently only one vsock device is supported. Calling this function
+ * multiple times will return an error.
+ *
+ * Arguments:
+ *  "ctx_id"       - the configuration context ID.
+ *  "tsi_features" - bitmask of TSI features (KRUN_TSI_HIJACK_INET, KRUN_TSI_HIJACK_UNIX)
+ *                   Use 0 to add vsock without any TSI hijacking.
+ *
+ * Returns:
+ *  Zero on success or a negative error number on failure.
+ */
+int32_t krun_add_vsock(uint32_t ctx_id, uint32_t tsi_features);
+
 /**
  * Returns the eventfd file descriptor to signal the guest to shut down orderly. This must be
  * called before starting the microVM with "krun_start_event". Only available in libkrun-efi.
@@ -955,6 +980,20 @@ int32_t krun_nitro_set_start_flags(uint32_t ctx_id, uint64_t start_flags);
  */
 int32_t krun_disable_implicit_console(uint32_t ctx_id);
 
+/**
+ * Disable the implicit vsock device.
+ *
+ * By default, libkrun creates a vsock device automatically. This function
+ * disables that behavior entirely - no vsock device will be created.
+ *
+ * Arguments:
+ *  "ctx_id" - the configuration context ID.
+ *
+ * Returns:
+ *  Zero on success or a negative error number on failure.
+ */
+int32_t krun_disable_implicit_vsock(uint32_t ctx_id);
+
 /*
  * Specify the value of `console=` in the kernel commandline.
  *

src/devices/src/virtio/vsock/device.rs

@@ -18,6 +18,7 @@ use super::super::{
 };
 use super::muxer::VsockMuxer;
 use super::packet::VsockPacket;
+use super::TsiFlags;
 use super::{defs, defs::uapi};
 use crate::virtio::InterruptTransport;
 
@@ -52,8 +53,7 @@ impl Vsock {
         host_port_map: Option<HashMap<u16, u16>>,
         queues: Vec<VirtQueue>,
         unix_ipc_port_map: Option<HashMap<u32, (PathBuf, bool)>>,
-        enable_tsi: bool,
-        enable_tsi_unix: bool,
+        tsi_flags: TsiFlags,
     ) -> super::Result<Vsock> {
         let mut queue_events = Vec::new();
         for _ in 0..queues.len() {
@@ -66,13 +66,7 @@ impl Vsock {
 
         Ok(Vsock {
             cid,
-            muxer: VsockMuxer::new(
-                cid,
-                host_port_map,
-                unix_ipc_port_map,
-                enable_tsi,
-                enable_tsi_unix,
-            ),
+            muxer: VsockMuxer::new(cid, host_port_map, unix_ipc_port_map, tsi_flags),
             queue_rx,
             queue_tx,
             queues,
@@ -90,21 +84,13 @@ impl Vsock {
         cid: u64,
         host_port_map: Option<HashMap<u16, u16>>,
         unix_ipc_port_map: Option<HashMap<u32, (PathBuf, bool)>>,
-        enable_tsi: bool,
-        enable_tsi_unix: bool,
+        tsi_flags: TsiFlags,
     ) -> super::Result<Vsock> {
         let queues: Vec<VirtQueue> = defs::QUEUE_SIZES
             .iter()
             .map(|&max_size| VirtQueue::new(max_size))
             .collect();
-        Self::with_queues(
-            cid,
-            host_port_map,
-            queues,
-            unix_ipc_port_map,
-            enable_tsi,
-            enable_tsi_unix,
-        )
+        Self::with_queues(cid, host_port_map, queues, unix_ipc_port_map, tsi_flags)
     }
 
     pub fn id(&self) -> &str {

src/devices/src/virtio/vsock/mod.rs

@@ -21,11 +21,39 @@ mod tsi_stream;
 mod unix;
 
 pub use self::defs::uapi::VIRTIO_ID_VSOCK as TYPE_VSOCK;
+pub use self::defs::TsiFlags;
 pub use self::device::Vsock;
 
+use bitflags::bitflags;
 use vm_memory::GuestMemoryError;
 
 mod defs {
+    use super::bitflags;
+
+    bitflags! {
+        /// TSI (Transparent Socket Impersonation) feature flags.
+        ///
+        /// These flags control which socket families are hijacked by TSI.
+        pub struct TsiFlags: u32 {
+            /// Hijack AF_INET and AF_INET6 sockets
+            const HIJACK_INET = 1 << 0;
+            /// Hijack AF_UNIX sockets
+            const HIJACK_UNIX = 1 << 1;
+        }
+    }
+
+    impl TsiFlags {
+        /// Returns true if any TSI hijacking is enabled.
+        pub fn tsi_enabled(&self) -> bool {
+            !self.is_empty()
+        }
+    }
+
+    impl Default for TsiFlags {
+        fn default() -> Self {
+            TsiFlags::empty()
+        }
+    }
     /// Device ID used in MMIO device identification.
     /// Because Vsock is unique per-vm, this ID can be hardcoded.
     pub const VSOCK_DEV_ID: &str = "vsock";

src/devices/src/virtio/vsock/muxer.rs

@@ -16,6 +16,7 @@ use super::timesync::TimesyncThread;
 use super::tsi_dgram::TsiDgramProxy;
 use super::tsi_stream::TsiStreamProxy;
 use super::unix::UnixProxy;
+use super::TsiFlags;
 use super::VsockError;
 use crossbeam_channel::{unbounded, Sender};
 use utils::epoll::{ControlOperation, Epoll, EpollEvent, EventSet};
@@ -106,17 +107,15 @@ pub struct VsockMuxer {
     proxy_map: ProxyMap,
     reaper_sender: Option<Sender<u64>>,
     unix_ipc_port_map: Option<HashMap<u32, (PathBuf, bool)>>,
-    enable_tsi: bool,
-    enable_tsi_unix: bool,
+    tsi_flags: TsiFlags,
 }
 
 impl VsockMuxer {
     pub(crate) fn new(
         cid: u64,
         host_port_map: Option<HashMap<u16, u16>>,
         unix_ipc_port_map: Option<HashMap<u32, (PathBuf, bool)>>,
-        enable_tsi: bool,
-        enable_tsi_unix: bool,
+        tsi_flags: TsiFlags,
     ) -> Self {
         VsockMuxer {
             cid,
@@ -129,8 +128,7 @@ impl VsockMuxer {
             proxy_map: Arc::new(RwLock::new(HashMap::new())),
             reaper_sender: None,
             unix_ipc_port_map,
-            enable_tsi,
-            enable_tsi_unix,
+            tsi_flags,
         }
     }
 
@@ -285,8 +283,16 @@ impl VsockMuxer {
                 defs::SOCK_STREAM => {
                     debug!("proxy create stream");
                     let id = ((req.peer_port as u64) << 32) | (defs::TSI_PROXY_PORT as u64);
-                    if req.family as i32 == libc::AF_UNIX && !self.enable_tsi_unix {
-                        warn!("rejecting tcp unix proxy because tsi_unix is disabled");
+                    if req.family as i32 == libc::AF_UNIX
+                        && !self.tsi_flags.contains(TsiFlags::HIJACK_UNIX)
+                    {
+                        warn!("rejecting stream unix proxy because HIJACK_UNIX is disabled");
+                        return;
+                    }
+                    if (req.family as i32 == libc::AF_INET || req.family as i32 == libc::AF_INET6)
+                        && !self.tsi_flags.contains(TsiFlags::HIJACK_INET)
+                    {
+                        warn!("rejecting stream inet proxy because HIJACK_INET is disabled");
                         return;
                     }
                     match TsiStreamProxy::new(
@@ -312,8 +318,16 @@ impl VsockMuxer {
                 defs::SOCK_DGRAM => {
                     debug!("proxy create dgram");
                     let id = ((req.peer_port as u64) << 32) | (defs::TSI_PROXY_PORT as u64);
-                    if req.family as i32 == libc::AF_UNIX && !self.enable_tsi_unix {
-                        warn!("rejecting udp unix proxy because tsi_unix is disabled");
+                    if req.family as i32 == libc::AF_UNIX
+                        && !self.tsi_flags.contains(TsiFlags::HIJACK_UNIX)
+                    {
+                        warn!("rejecting dgram unix proxy because HIJACK_UNIX is disabled");
+                        return;
+                    }
+                    if (req.family as i32 == libc::AF_INET || req.family as i32 == libc::AF_INET6)
+                        && !self.tsi_flags.contains(TsiFlags::HIJACK_INET)
+                    {
+                        warn!("rejecting dgram inet proxy because HIJACK_INET is disabled");
                         return;
                     }
                     match TsiDgramProxy::new(
@@ -498,14 +512,18 @@ impl VsockMuxer {
         }
 
         match pkt.dst_port() {
-            defs::TSI_PROXY_CREATE if self.enable_tsi => self.process_proxy_create(pkt),
-            defs::TSI_CONNECT if self.enable_tsi => self.process_connect(pkt),
-            defs::TSI_GETNAME if self.enable_tsi => self.process_getname(pkt),
-            defs::TSI_SENDTO_ADDR if self.enable_tsi => self.process_sendto_addr(pkt),
-            defs::TSI_SENDTO_DATA if self.enable_tsi => self.process_sendto_data(pkt),
-            defs::TSI_LISTEN if self.enable_tsi => self.process_listen_request(pkt),
-            defs::TSI_ACCEPT if self.enable_tsi => self.process_accept_request(pkt),
-            defs::TSI_PROXY_RELEASE if self.enable_tsi => self.process_proxy_release(pkt),
+            defs::TSI_PROXY_CREATE if self.tsi_flags.tsi_enabled() => {
+                self.process_proxy_create(pkt)
+            }
+            defs::TSI_CONNECT if self.tsi_flags.tsi_enabled() => self.process_connect(pkt),
+            defs::TSI_GETNAME if self.tsi_flags.tsi_enabled() => self.process_getname(pkt),
+            defs::TSI_SENDTO_ADDR if self.tsi_flags.tsi_enabled() => self.process_sendto_addr(pkt),
+            defs::TSI_SENDTO_DATA if self.tsi_flags.tsi_enabled() => self.process_sendto_data(pkt),
+            defs::TSI_LISTEN if self.tsi_flags.tsi_enabled() => self.process_listen_request(pkt),
+            defs::TSI_ACCEPT if self.tsi_flags.tsi_enabled() => self.process_accept_request(pkt),
+            defs::TSI_PROXY_RELEASE if self.tsi_flags.tsi_enabled() => {
+                self.process_proxy_release(pkt)
+            }
             _ => {
                 if pkt.op() == uapi::VSOCK_OP_RW {
                     self.process_dgram_rw(pkt);