1 // Copyright 2019 Intel Corporation. All Rights Reserved. 2 // SPDX-License-Identifier: Apache-2.0 3 4 use crate::seccomp_filters::Thread; 5 use crate::thread_helper::spawn_virtio_thread; 6 use crate::vhost_user::vu_common_ctrl::{VhostUserConfig, VhostUserHandle}; 7 use crate::vhost_user::{Error, Result, VhostUserCommon}; 8 use crate::{ 9 ActivateResult, NetCtrlEpollHandler, VirtioCommon, VirtioDevice, VirtioDeviceType, 10 VirtioInterrupt, VIRTIO_F_IOMMU_PLATFORM, VIRTIO_F_RING_EVENT_IDX, VIRTIO_F_VERSION_1, 11 }; 12 use crate::{GuestMemoryMmap, GuestRegionMmap}; 13 use net_util::{build_net_config_space, CtrlQueue, MacAddr, VirtioNetConfig}; 14 use seccompiler::SeccompAction; 15 use std::result; 16 use std::sync::atomic::AtomicBool; 17 use std::sync::{Arc, Barrier, Mutex}; 18 use std::thread; 19 use std::vec::Vec; 20 use versionize::{VersionMap, Versionize, VersionizeResult}; 21 use versionize_derive::Versionize; 22 use vhost::vhost_user::message::{VhostUserProtocolFeatures, VhostUserVirtioFeatures}; 23 use vhost::vhost_user::{MasterReqHandler, VhostUserMaster, VhostUserMasterReqHandler}; 24 use virtio_bindings::bindings::virtio_net::{ 25 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_CTRL_VQ, VIRTIO_NET_F_GUEST_CSUM, VIRTIO_NET_F_GUEST_ECN, 26 VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, VIRTIO_NET_F_GUEST_UFO, 27 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_TSO6, VIRTIO_NET_F_HOST_UFO, 28 VIRTIO_NET_F_MAC, VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_MTU, 29 }; 30 use virtio_bindings::bindings::virtio_ring::VIRTIO_RING_F_EVENT_IDX; 31 use virtio_queue::{Queue, QueueT}; 32 use vm_memory::{ByteValued, GuestMemoryAtomic}; 33 use vm_migration::{ 34 protocol::MemoryRangeTable, Migratable, MigratableError, Pausable, Snapshot, Snapshottable, 35 Transportable, VersionMapped, 36 }; 37 use vmm_sys_util::eventfd::EventFd; 38 39 const DEFAULT_QUEUE_NUMBER: usize = 2; 40 41 #[derive(Versionize)] 42 pub struct State { 43 pub avail_features: u64, 44 pub acked_features: u64, 45 pub config: VirtioNetConfig, 46 pub acked_protocol_features: u64, 47 pub vu_num_queues: usize, 48 } 49 50 impl VersionMapped for State {} 51 52 struct SlaveReqHandler {} 53 impl VhostUserMasterReqHandler for SlaveReqHandler {} 54 55 pub struct Net { 56 common: VirtioCommon, 57 vu_common: VhostUserCommon, 58 id: String, 59 config: VirtioNetConfig, 60 guest_memory: Option<GuestMemoryAtomic<GuestMemoryMmap>>, 61 ctrl_queue_epoll_thread: Option<thread::JoinHandle<()>>, 62 epoll_thread: Option<thread::JoinHandle<()>>, 63 seccomp_action: SeccompAction, 64 exit_evt: EventFd, 65 iommu: bool, 66 } 67 68 impl Net { 69 /// Create a new vhost-user-net device 70 #[allow(clippy::too_many_arguments)] 71 pub fn new( 72 id: String, 73 mac_addr: MacAddr, 74 mtu: Option<u16>, 75 vu_cfg: VhostUserConfig, 76 server: bool, 77 seccomp_action: SeccompAction, 78 exit_evt: EventFd, 79 iommu: bool, 80 state: Option<State>, 81 ) -> Result<Net> { 82 let mut num_queues = vu_cfg.num_queues; 83 84 let mut vu = 85 VhostUserHandle::connect_vhost_user(server, &vu_cfg.socket, num_queues as u64, false)?; 86 87 let ( 88 avail_features, 89 acked_features, 90 acked_protocol_features, 91 vu_num_queues, 92 config, 93 paused, 94 ) = if let Some(state) = state { 95 info!("Restoring vhost-user-net {}", id); 96 97 // The backend acknowledged features must not contain 98 // VIRTIO_NET_F_MAC since we don't expect the backend 99 // to handle it. 100 let backend_acked_features = state.acked_features & !(1 << VIRTIO_NET_F_MAC); 101 102 vu.set_protocol_features_vhost_user( 103 backend_acked_features, 104 state.acked_protocol_features, 105 )?; 106 107 // If the control queue feature has been negotiated, let's 108 // increase the number of queues. 109 if state.acked_features & (1 << VIRTIO_NET_F_CTRL_VQ) != 0 { 110 num_queues += 1; 111 } 112 113 ( 114 state.avail_features, 115 state.acked_features, 116 state.acked_protocol_features, 117 state.vu_num_queues, 118 state.config, 119 true, 120 ) 121 } else { 122 // Filling device and vring features VMM supports. 123 let mut avail_features = 1 << VIRTIO_NET_F_CSUM 124 | 1 << VIRTIO_NET_F_GUEST_CSUM 125 | 1 << VIRTIO_NET_F_GUEST_TSO4 126 | 1 << VIRTIO_NET_F_GUEST_TSO6 127 | 1 << VIRTIO_NET_F_GUEST_ECN 128 | 1 << VIRTIO_NET_F_GUEST_UFO 129 | 1 << VIRTIO_NET_F_HOST_TSO4 130 | 1 << VIRTIO_NET_F_HOST_TSO6 131 | 1 << VIRTIO_NET_F_HOST_ECN 132 | 1 << VIRTIO_NET_F_HOST_UFO 133 | 1 << VIRTIO_NET_F_MRG_RXBUF 134 | 1 << VIRTIO_NET_F_CTRL_VQ 135 | 1 << VIRTIO_F_RING_EVENT_IDX 136 | 1 << VIRTIO_F_VERSION_1 137 | VhostUserVirtioFeatures::PROTOCOL_FEATURES.bits(); 138 139 if mtu.is_some() { 140 avail_features |= 1u64 << VIRTIO_NET_F_MTU; 141 } 142 143 let mut config = VirtioNetConfig::default(); 144 build_net_config_space(&mut config, mac_addr, num_queues, mtu, &mut avail_features); 145 146 let avail_protocol_features = VhostUserProtocolFeatures::MQ 147 | VhostUserProtocolFeatures::CONFIGURE_MEM_SLOTS 148 | VhostUserProtocolFeatures::REPLY_ACK 149 | VhostUserProtocolFeatures::INFLIGHT_SHMFD 150 | VhostUserProtocolFeatures::LOG_SHMFD; 151 152 let (mut acked_features, acked_protocol_features) = 153 vu.negotiate_features_vhost_user(avail_features, avail_protocol_features)?; 154 155 let backend_num_queues = 156 if acked_protocol_features & VhostUserProtocolFeatures::MQ.bits() != 0 { 157 vu.socket_handle() 158 .get_queue_num() 159 .map_err(Error::VhostUserGetQueueMaxNum)? as usize 160 } else { 161 DEFAULT_QUEUE_NUMBER 162 }; 163 164 if num_queues > backend_num_queues { 165 error!("vhost-user-net requested too many queues ({}) since the backend only supports {}\n", 166 num_queues, backend_num_queues); 167 return Err(Error::BadQueueNum); 168 } 169 170 // If the control queue feature has been negotiated, let's increase 171 // the number of queues. 172 let vu_num_queues = num_queues; 173 if acked_features & (1 << VIRTIO_NET_F_CTRL_VQ) != 0 { 174 num_queues += 1; 175 } 176 177 // Make sure the virtio feature to set the MAC address is exposed to 178 // the guest, even if it hasn't been negotiated with the backend. 179 acked_features |= 1 << VIRTIO_NET_F_MAC; 180 181 ( 182 acked_features, 183 // If part of the available features that have been acked, 184 // the PROTOCOL_FEATURES bit must be already set through 185 // the VIRTIO acked features as we know the guest would 186 // never ack it, thus the feature would be lost. 187 acked_features & VhostUserVirtioFeatures::PROTOCOL_FEATURES.bits(), 188 acked_protocol_features, 189 vu_num_queues, 190 config, 191 false, 192 ) 193 }; 194 195 Ok(Net { 196 id, 197 common: VirtioCommon { 198 device_type: VirtioDeviceType::Net as u32, 199 queue_sizes: vec![vu_cfg.queue_size; num_queues], 200 avail_features, 201 acked_features, 202 paused_sync: Some(Arc::new(Barrier::new(2))), 203 min_queues: DEFAULT_QUEUE_NUMBER as u16, 204 paused: Arc::new(AtomicBool::new(paused)), 205 ..Default::default() 206 }, 207 vu_common: VhostUserCommon { 208 vu: Some(Arc::new(Mutex::new(vu))), 209 acked_protocol_features, 210 socket_path: vu_cfg.socket, 211 vu_num_queues, 212 server, 213 ..Default::default() 214 }, 215 config, 216 guest_memory: None, 217 ctrl_queue_epoll_thread: None, 218 epoll_thread: None, 219 seccomp_action, 220 exit_evt, 221 iommu, 222 }) 223 } 224 225 fn state(&self) -> State { 226 State { 227 avail_features: self.common.avail_features, 228 acked_features: self.common.acked_features, 229 config: self.config, 230 acked_protocol_features: self.vu_common.acked_protocol_features, 231 vu_num_queues: self.vu_common.vu_num_queues, 232 } 233 } 234 } 235 236 impl Drop for Net { 237 fn drop(&mut self) { 238 if let Some(kill_evt) = self.common.kill_evt.take() { 239 if let Err(e) = kill_evt.write(1) { 240 error!("failed to kill vhost-user-net: {:?}", e); 241 } 242 } 243 } 244 } 245 246 impl VirtioDevice for Net { 247 fn device_type(&self) -> u32 { 248 self.common.device_type 249 } 250 251 fn queue_max_sizes(&self) -> &[u16] { 252 &self.common.queue_sizes 253 } 254 255 fn features(&self) -> u64 { 256 let mut features = self.common.avail_features; 257 if self.iommu { 258 features |= 1u64 << VIRTIO_F_IOMMU_PLATFORM; 259 } 260 features 261 } 262 263 fn ack_features(&mut self, value: u64) { 264 self.common.ack_features(value) 265 } 266 267 fn read_config(&self, offset: u64, data: &mut [u8]) { 268 self.read_config_from_slice(self.config.as_slice(), offset, data); 269 } 270 271 fn activate( 272 &mut self, 273 mem: GuestMemoryAtomic<GuestMemoryMmap>, 274 interrupt_cb: Arc<dyn VirtioInterrupt>, 275 mut queues: Vec<(usize, Queue, EventFd)>, 276 ) -> ActivateResult { 277 self.common.activate(&queues, &interrupt_cb)?; 278 self.guest_memory = Some(mem.clone()); 279 280 let num_queues = queues.len(); 281 let event_idx = self.common.feature_acked(VIRTIO_RING_F_EVENT_IDX.into()); 282 if self.common.feature_acked(VIRTIO_NET_F_CTRL_VQ.into()) && num_queues % 2 != 0 { 283 let ctrl_queue_index = num_queues - 1; 284 let (_, mut ctrl_queue, ctrl_queue_evt) = queues.remove(ctrl_queue_index); 285 286 ctrl_queue.set_event_idx(event_idx); 287 288 let (kill_evt, pause_evt) = self.common.dup_eventfds(); 289 290 let mut ctrl_handler = NetCtrlEpollHandler { 291 mem: mem.clone(), 292 kill_evt, 293 pause_evt, 294 ctrl_q: CtrlQueue::new(Vec::new()), 295 queue: ctrl_queue, 296 queue_evt: ctrl_queue_evt, 297 access_platform: None, 298 interrupt_cb: interrupt_cb.clone(), 299 queue_index: ctrl_queue_index as u16, 300 }; 301 302 let paused = self.common.paused.clone(); 303 // Let's update the barrier as we need 1 for the control queue 304 // thread + 1 for the common vhost-user thread + 1 for the main 305 // thread signalling the pause. 306 self.common.paused_sync = Some(Arc::new(Barrier::new(3))); 307 let paused_sync = self.common.paused_sync.clone(); 308 309 let mut epoll_threads = Vec::new(); 310 spawn_virtio_thread( 311 &format!("{}_ctrl", &self.id), 312 &self.seccomp_action, 313 Thread::VirtioVhostNetCtl, 314 &mut epoll_threads, 315 &self.exit_evt, 316 move || ctrl_handler.run_ctrl(paused, paused_sync.unwrap()), 317 )?; 318 self.ctrl_queue_epoll_thread = Some(epoll_threads.remove(0)); 319 } 320 321 let slave_req_handler: Option<MasterReqHandler<SlaveReqHandler>> = None; 322 323 // The backend acknowledged features must not contain VIRTIO_NET_F_MAC 324 // since we don't expect the backend to handle it. 325 let backend_acked_features = self.common.acked_features & !(1 << VIRTIO_NET_F_MAC); 326 327 // Run a dedicated thread for handling potential reconnections with 328 // the backend. 329 let (kill_evt, pause_evt) = self.common.dup_eventfds(); 330 331 let mut handler = self.vu_common.activate( 332 mem, 333 queues, 334 interrupt_cb, 335 backend_acked_features, 336 slave_req_handler, 337 kill_evt, 338 pause_evt, 339 )?; 340 341 let paused = self.common.paused.clone(); 342 let paused_sync = self.common.paused_sync.clone(); 343 344 let mut epoll_threads = Vec::new(); 345 spawn_virtio_thread( 346 &self.id, 347 &self.seccomp_action, 348 Thread::VirtioVhostNet, 349 &mut epoll_threads, 350 &self.exit_evt, 351 move || handler.run(paused, paused_sync.unwrap()), 352 )?; 353 self.epoll_thread = Some(epoll_threads.remove(0)); 354 355 Ok(()) 356 } 357 358 fn reset(&mut self) -> Option<Arc<dyn VirtioInterrupt>> { 359 // We first must resume the virtio thread if it was paused. 360 if self.common.pause_evt.take().is_some() { 361 self.common.resume().ok()?; 362 } 363 364 if let Some(vu) = &self.vu_common.vu { 365 if let Err(e) = vu.lock().unwrap().reset_vhost_user() { 366 error!("Failed to reset vhost-user daemon: {:?}", e); 367 return None; 368 } 369 } 370 371 if let Some(kill_evt) = self.common.kill_evt.take() { 372 // Ignore the result because there is nothing we can do about it. 373 let _ = kill_evt.write(1); 374 } 375 376 event!("virtio-device", "reset", "id", &self.id); 377 378 // Return the interrupt 379 Some(self.common.interrupt_cb.take().unwrap()) 380 } 381 382 fn shutdown(&mut self) { 383 self.vu_common.shutdown(); 384 } 385 386 fn add_memory_region( 387 &mut self, 388 region: &Arc<GuestRegionMmap>, 389 ) -> std::result::Result<(), crate::Error> { 390 self.vu_common.add_memory_region(&self.guest_memory, region) 391 } 392 } 393 394 impl Pausable for Net { 395 fn pause(&mut self) -> result::Result<(), MigratableError> { 396 self.vu_common.pause()?; 397 self.common.pause() 398 } 399 400 fn resume(&mut self) -> result::Result<(), MigratableError> { 401 self.common.resume()?; 402 403 if let Some(epoll_thread) = &self.epoll_thread { 404 epoll_thread.thread().unpark(); 405 } 406 407 if let Some(ctrl_queue_epoll_thread) = &self.ctrl_queue_epoll_thread { 408 ctrl_queue_epoll_thread.thread().unpark(); 409 } 410 411 self.vu_common.resume() 412 } 413 } 414 415 impl Snapshottable for Net { 416 fn id(&self) -> String { 417 self.id.clone() 418 } 419 420 fn snapshot(&mut self) -> std::result::Result<Snapshot, MigratableError> { 421 self.vu_common.snapshot(&self.state()) 422 } 423 } 424 impl Transportable for Net {} 425 426 impl Migratable for Net { 427 fn start_dirty_log(&mut self) -> std::result::Result<(), MigratableError> { 428 self.vu_common.start_dirty_log(&self.guest_memory) 429 } 430 431 fn stop_dirty_log(&mut self) -> std::result::Result<(), MigratableError> { 432 self.vu_common.stop_dirty_log() 433 } 434 435 fn dirty_log(&mut self) -> std::result::Result<MemoryRangeTable, MigratableError> { 436 self.vu_common.dirty_log(&self.guest_memory) 437 } 438 439 fn start_migration(&mut self) -> std::result::Result<(), MigratableError> { 440 self.vu_common.start_migration() 441 } 442 443 fn complete_migration(&mut self) -> std::result::Result<(), MigratableError> { 444 self.vu_common 445 .complete_migration(self.common.kill_evt.take()) 446 } 447 } 448