virtio-devices: vhost_user: blk: Support socket reconnection handling

This commit enables socket reconnection for vhost-user-blk backends.

Note that, till this commit, inflight I/O trakcing and restoring is not
supported. Therefore, only vhost-user-blk backend that are not processing
inflight requests can work normally after reconnection.

Signed-off-by: Jiachen Zhang <zhangjiachen.jaycee@bytedance.com>

virtio-devices/src/vhost_user/blk.rs

@@ -5,10 +5,11 @@ use super::super::{
     ActivateError, ActivateResult, Queue, VirtioCommon, VirtioDevice, VirtioDeviceType,
 };
 use super::vu_common_ctrl::{
-    add_memory_region, negotiate_features_vhost_user, reset_vhost_user, setup_vhost_user,
-    update_mem_table, VhostUserConfig,
+    add_memory_region, connect_vhost_user, negotiate_features_vhost_user, reset_vhost_user,
+    setup_vhost_user, update_mem_table, VhostUserConfig,
 };
 use super::{Error, Result, DEFAULT_VIRTIO_FEATURES};
+use crate::vhost_user::ReconnectEpollHandler;
 use crate::VirtioInterrupt;
 use crate::{GuestMemoryMmap, GuestRegionMmap};
 use block_util::VirtioBlockConfig;
@@ -16,7 +17,8 @@ use std::mem;
 use std::ops::Deref;
 use std::os::unix::io::AsRawFd;
 use std::result;
-use std::sync::{Arc, Barrier};
+use std::sync::{Arc, Barrier, Mutex};
+use std::thread;
 use std::vec::Vec;
 use vhost::vhost_user::message::VhostUserConfigFlags;
 use vhost::vhost_user::message::VHOST_USER_CONFIG_OFFSET;
@@ -40,10 +42,12 @@ impl VhostUserMasterReqHandler for SlaveReqHandler {}
 pub struct Blk {
     common: VirtioCommon,
     id: String,
-    vhost_user_blk: Master,
+    vhost_user_blk: Arc<Mutex<Master>>,
     config: VirtioBlockConfig,
     guest_memory: Option<GuestMemoryAtomic<GuestMemoryMmap>>,
     acked_protocol_features: u64,
+    socket_path: String,
+    reconnect_epoll_thread: Option<thread::JoinHandle<()>>,
 }
 
 impl Blk {
@@ -51,8 +55,8 @@ impl Blk {
     pub fn new(id: String, vu_cfg: VhostUserConfig) -> Result<Blk> {
         let num_queues = vu_cfg.num_queues;
 
-        let mut vhost_user_blk = Master::connect(&vu_cfg.socket, num_queues as u64)
-            .map_err(Error::VhostUserCreateMaster)?;
+        let mut vhost_user_blk =
+            connect_vhost_user(false, &vu_cfg.socket, num_queues as u64, false)?;
 
         // Filling device and vring features VMM supports.
         let mut avail_features = 1 << VIRTIO_BLK_F_SIZE_MAX
@@ -128,15 +132,17 @@ impl Blk {
                 queue_sizes: vec![vu_cfg.queue_size; num_queues],
                 avail_features: acked_features,
                 acked_features: 0,
-                paused_sync: Some(Arc::new(Barrier::new(1))),
+                paused_sync: Some(Arc::new(Barrier::new(2))),
                 min_queues: DEFAULT_QUEUE_NUMBER as u16,
                 ..Default::default()
             },
             id,
-            vhost_user_blk,
+            vhost_user_blk: Arc::new(Mutex::new(vhost_user_blk)),
             config,
             guest_memory: None,
             acked_protocol_features,
+            socket_path: vu_cfg.socket,
+            reconnect_epoll_thread: None,
         })
     }
 }
@@ -189,6 +195,8 @@ impl VirtioDevice for Blk {
         self.config.writeback = data[0];
         if let Err(e) = self
             .vhost_user_blk
+            .lock()
+            .unwrap()
             .set_config(offset as u32, VhostUserConfigFlags::WRITABLE, data)
             .map_err(Error::VhostUserSetConfig)
         {
@@ -207,17 +215,56 @@ impl VirtioDevice for Blk {
 
         self.guest_memory = Some(mem.clone());
 
+        // The backend acknowledged features must contain the protocol feature
+        // bit in case it was initially set but lost through the features
+        // negotiation with the guest.
+        let backend_acked_features = self.common.acked_features
+            | (self.common.avail_features & VhostUserVirtioFeatures::PROTOCOL_FEATURES.bits());
+
         setup_vhost_user(
-            &mut self.vhost_user_blk,
+            &mut self.vhost_user_blk.lock().unwrap(),
             &mem.memory(),
-            queues,
-            queue_evts,
+            queues.clone(),
+            queue_evts.iter().map(|q| q.try_clone().unwrap()).collect(),
             &interrupt_cb,
-            self.common.acked_features
-                | (self.common.avail_features & VhostUserVirtioFeatures::PROTOCOL_FEATURES.bits()),
+            backend_acked_features,
         )
         .map_err(ActivateError::VhostUserBlkSetup)?;
 
+        // Run a dedicated thread for handling potential reconnections with
+        // the backend.
+        let (kill_evt, pause_evt) = self.common.dup_eventfds();
+
+        let mut reconnect_handler = ReconnectEpollHandler {
+            vu: self.vhost_user_blk.clone(),
+            mem,
+            kill_evt,
+            pause_evt,
+            queues,
+            queue_evts,
+            virtio_interrupt: interrupt_cb,
+            acked_features: backend_acked_features,
+            acked_protocol_features: self.acked_protocol_features,
+            socket_path: self.socket_path.clone(),
+            server: false,
+        };
+
+        let paused = self.common.paused.clone();
+        let paused_sync = self.common.paused_sync.clone();
+
+        thread::Builder::new()
+            .name(format!("{}_reconnect", self.id))
+            .spawn(move || {
+                if let Err(e) = reconnect_handler.run(paused, paused_sync.unwrap()) {
+                    error!("Error running reconnection worker: {:?}", e);
+                }
+            })
+            .map(|thread| self.reconnect_epoll_thread = Some(thread))
+            .map_err(|e| {
+                error!("failed to clone queue EventFd: {}", e);
+                ActivateError::BadActivate
+            })?;
+
         Ok(())
     }
 
@@ -227,7 +274,10 @@ impl VirtioDevice for Blk {
             self.common.resume().ok()?;
         }
 
-        if let Err(e) = reset_vhost_user(&mut self.vhost_user_blk, self.common.queue_sizes.len()) {
+        if let Err(e) = reset_vhost_user(
+            &mut self.vhost_user_blk.lock().unwrap(),
+            self.common.queue_sizes.len(),
+        ) {
             error!("Failed to reset vhost-user daemon: {:?}", e);
             return None;
         }
@@ -244,7 +294,7 @@ impl VirtioDevice for Blk {
     }
 
     fn shutdown(&mut self) {
-        let _ = unsafe { libc::close(self.vhost_user_blk.as_raw_fd()) };
+        let _ = unsafe { libc::close(self.vhost_user_blk.lock().unwrap().as_raw_fd()) };
     }
 
     fn add_memory_region(
@@ -253,11 +303,14 @@ impl VirtioDevice for Blk {
     ) -> std::result::Result<(), crate::Error> {
         if self.acked_protocol_features & VhostUserProtocolFeatures::CONFIGURE_MEM_SLOTS.bits() != 0
         {
-            add_memory_region(&mut self.vhost_user_blk, region)
+            add_memory_region(&mut self.vhost_user_blk.lock().unwrap(), region)
                 .map_err(crate::Error::VhostUserAddMemoryRegion)
         } else if let Some(guest_memory) = &self.guest_memory {
-            update_mem_table(&mut self.vhost_user_blk, guest_memory.memory().deref())
-                .map_err(crate::Error::VhostUserUpdateMemory)
+            update_mem_table(
+                &mut self.vhost_user_blk.lock().unwrap(),
+                guest_memory.memory().deref(),
+            )
+            .map_err(crate::Error::VhostUserUpdateMemory)
         } else {
             Ok(())
         }
@@ -270,7 +323,12 @@ impl Pausable for Blk {
     }
 
     fn resume(&mut self) -> result::Result<(), MigratableError> {
-        self.common.resume()
+        self.common.resume()?;
+
+        if let Some(reconnect_epoll_thread) = &self.reconnect_epoll_thread {
+            reconnect_epoll_thread.thread().unpark();
+        }
+        Ok(())
     }
 }