runtime-rs: block volume

support block volume in runtime-rs

Fixes: #5375
Signed-off-by: Zhongtao Hu <zhongtaohu.tim@linux.alibaba.com>
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>

src/runtime-rs/crates/resource/src/manager_inner.rs

@@ -236,7 +236,13 @@ impl ResourceManagerInner {
         spec: &oci::Spec,
     ) -> Result<Vec<Arc<dyn Volume>>> {
         self.volume_resource
-            .handler_volumes(&self.share_fs, cid, spec)
+            .handler_volumes(
+                &self.share_fs,
+                cid,
+                spec,
+                self.device_manager.as_ref(),
+                &self.sid,
+            )
             .await
     }
 

src/runtime-rs/crates/resource/src/volume/block_volume.rs

@@ -6,27 +6,126 @@
 
 use anyhow::Result;
 use async_trait::async_trait;
+use std::{collections::HashMap, fs, path::Path};
 
-use super::Volume;
+use crate::share_fs::{do_get_guest_path, do_get_host_path};
 
+use super::{share_fs_volume::generate_mount_path, Volume};
+use agent::Storage;
+use anyhow::{anyhow, Context};
+use hypervisor::{device::DeviceManager, BlockConfig, DeviceConfig};
+use nix::sys::stat::{self, SFlag};
+use tokio::sync::RwLock;
 #[derive(Debug)]
-pub(crate) struct BlockVolume {}
+pub(crate) struct BlockVolume {
+    storage: Option<agent::Storage>,
+    mount: oci::Mount,
+    device_id: String,
+}
 
 /// BlockVolume: block device volume
 impl BlockVolume {
-    pub(crate) fn new(_m: &oci::Mount) -> Result<Self> {
-        Ok(Self {})
+    pub(crate) async fn new(
+        d: &RwLock<DeviceManager>,
+        m: &oci::Mount,
+        read_only: bool,
+        cid: &str,
+        sid: &str,
+    ) -> Result<Self> {
+        let fstat = stat::stat(m.source.as_str()).context(format!("stat {}", m.source))?;
+        info!(sl!(), "device stat: {:?}", fstat);
+        let mut options = HashMap::new();
+        if read_only {
+            options.insert("read_only".to_string(), "true".to_string());
+        }
+
+        let block_device_config = &mut BlockConfig {
+            major: stat::major(fstat.st_rdev) as i64,
+            minor: stat::minor(fstat.st_rdev) as i64,
+            ..Default::default()
+        };
+
+        let device_id = d
+            .write()
+            .await
+            .new_device(&DeviceConfig::Block(block_device_config.clone()))
+            .await
+            .context("failed to create deviec")?;
+
+        d.write()
+            .await
+            .try_add_device(device_id.as_str())
+            .await
+            .context("failed to add deivce")?;
+
+        let file_name = Path::new(&m.source).file_name().unwrap().to_str().unwrap();
+        let file_name = generate_mount_path(cid, file_name);
+        let guest_path = do_get_guest_path(&file_name, cid, true, false);
+        let host_path = do_get_host_path(&file_name, sid, cid, true, read_only);
+        fs::create_dir_all(&host_path)
+            .map_err(|e| anyhow!("failed to create rootfs dir {}: {:?}", host_path, e))?;
+
+        // get complete device information
+        let dev_info = d
+            .read()
+            .await
+            .get_device_info(&device_id)
+            .await
+            .context("failed to get device info")?;
+
+        // storage
+        let mut storage = Storage::default();
+
+        if let DeviceConfig::Block(config) = dev_info {
+            storage.driver = config.driver_option;
+            storage.source = config.virt_path;
+        }
+
+        storage.options = if read_only {
+            vec!["ro".to_string()]
+        } else {
+            Vec::new()
+        };
+
+        storage.mount_point = guest_path.clone();
+
+        // If the volume had specified the filesystem type, use it. Otherwise, set it
+        // to ext4 since but right now we only support it.
+        if m.r#type != "bind" {
+            storage.fs_type = m.r#type.clone();
+        } else {
+            storage.fs_type = "ext4".to_string();
+        }
+
+        // mount
+        let mount = oci::Mount {
+            destination: m.destination.clone(),
+            r#type: m.r#type.clone(),
+            source: guest_path.clone(),
+            options: m.options.clone(),
+        };
+
+        Ok(Self {
+            storage: Some(storage),
+            mount,
+            device_id,
+        })
     }
 }
 
 #[async_trait]
 impl Volume for BlockVolume {
-    fn get_volume_mount(&self) -> anyhow::Result<Vec<oci::Mount>> {
-        todo!()
+    fn get_volume_mount(&self) -> Result<Vec<oci::Mount>> {
+        Ok(vec![self.mount.clone()])
     }
 
     fn get_storage(&self) -> Result<Vec<agent::Storage>> {
-        todo!()
+        let s = if let Some(s) = self.storage.as_ref() {
+            vec![s.clone()]
+        } else {
+            vec![]
+        };
+        Ok(s)
     }
 
     async fn cleanup(&self) -> Result<()> {
@@ -34,9 +133,19 @@ impl Volume for BlockVolume {
         warn!(sl!(), "Cleaning up BlockVolume is still unimplemented.");
         Ok(())
     }
+
+    fn get_device_id(&self) -> Result<Option<String>> {
+        Ok(Some(self.device_id.clone()))
+    }
 }
 
-pub(crate) fn is_block_volume(_m: &oci::Mount) -> bool {
-    // attach block device
+pub(crate) fn is_block_volume(m: &oci::Mount) -> bool {
+    if m.r#type != "bind" {
+        return false;
+    }
+    if let Ok(fstat) = stat::stat(m.source.as_str()).context(format!("stat {}", m.source)) {
+        info!(sl!(), "device stat: {:?}", fstat);
+        return SFlag::from_bits_truncate(fstat.st_mode) == SFlag::S_IFBLK;
+    }
     false
 }

src/runtime-rs/crates/resource/src/volume/default_volume.rs

@@ -38,4 +38,8 @@ impl Volume for DefaultVolume {
         warn!(sl!(), "Cleaning up DefaultVolume is still unimplemented.");
         Ok(())
     }
+
+    fn get_device_id(&self) -> Result<Option<String>> {
+        Ok(None)
+    }
 }

src/runtime-rs/crates/resource/src/volume/hugepage.rs

@@ -91,6 +91,10 @@ impl Volume for Hugepage {
     async fn cleanup(&self) -> Result<()> {
         Ok(())
     }
+
+    fn get_device_id(&self) -> Result<Option<String>> {
+        Ok(None)
+    }
 }
 
 pub(crate) fn get_huge_page_option(m: &oci::Mount) -> Result<Option<Vec<String>>> {

src/runtime-rs/crates/resource/src/volume/mod.rs

@@ -12,10 +12,11 @@ mod shm_volume;
 use async_trait::async_trait;
 
 use anyhow::{Context, Result};
+use hypervisor::device::DeviceManager;
 use std::{sync::Arc, vec::Vec};
 use tokio::sync::RwLock;
 
-use crate::share_fs::ShareFs;
+use crate::{share_fs::ShareFs, volume::block_volume::is_block_volume};
 
 use self::hugepage::{get_huge_page_limits_map, get_huge_page_option};
 
@@ -25,6 +26,7 @@ const BIND: &str = "bind";
 pub trait Volume: Send + Sync {
     fn get_volume_mount(&self) -> Result<Vec<oci::Mount>>;
     fn get_storage(&self) -> Result<Vec<agent::Storage>>;
+    fn get_device_id(&self) -> Result<Option<String>>;
     async fn cleanup(&self) -> Result<()>;
 }
 
@@ -48,20 +50,25 @@ impl VolumeResource {
         share_fs: &Option<Arc<dyn ShareFs>>,
         cid: &str,
         spec: &oci::Spec,
+        d: &RwLock<DeviceManager>,
+        sid: &str,
     ) -> Result<Vec<Arc<dyn Volume>>> {
         let mut volumes: Vec<Arc<dyn Volume>> = vec![];
         let oci_mounts = &spec.mounts;
+        info!(sl!(), " oci mount is : {:?}", oci_mounts.clone());
         // handle mounts
         for m in oci_mounts {
+            let read_only = m.options.iter().any(|opt| opt == "ro");
             let volume: Arc<dyn Volume> = if shm_volume::is_shim_volume(m) {
                 let shm_size = shm_volume::DEFAULT_SHM_SIZE;
                 Arc::new(
                     shm_volume::ShmVolume::new(m, shm_size)
                         .with_context(|| format!("new shm volume {:?}", m))?,
                 )
-            } else if share_fs_volume::is_share_fs_volume(m) {
+            } else if is_block_volume(m) {
+                // handle block volume
                 Arc::new(
-                    share_fs_volume::ShareFsVolume::new(share_fs, m, cid)
+                    block_volume::BlockVolume::new(d, m, read_only, cid, sid)
                         .await
                         .with_context(|| format!("new share fs volume {:?}", m))?,
                 )
@@ -76,10 +83,11 @@ impl VolumeResource {
                     hugepage::Hugepage::new(m, hugepage_limits, options)
                         .with_context(|| format!("handle hugepages {:?}", m))?,
                 )
-            } else if block_volume::is_block_volume(m) {
+            } else if share_fs_volume::is_share_fs_volume(m) {
                 Arc::new(
-                    block_volume::BlockVolume::new(m)
-                        .with_context(|| format!("new block volume {:?}", m))?,
+                    share_fs_volume::ShareFsVolume::new(share_fs, m, cid, read_only)
+                        .await
+                        .with_context(|| format!("new share fs volume {:?}", m))?,
                 )
             } else if is_skip_volume(m) {
                 info!(sl!(), "skip volume {:?}", m);

src/runtime-rs/crates/resource/src/volume/share_fs_volume.rs

@@ -36,6 +36,7 @@ impl ShareFsVolume {
         share_fs: &Option<Arc<dyn ShareFs>>,
         m: &oci::Mount,
         cid: &str,
+        readonly: bool,
     ) -> Result<Self> {
         // The file_name is in the format of "sandbox-{uuid}-{file_name}"
         let file_name = Path::new(&m.source).file_name().unwrap().to_str().unwrap();
@@ -69,8 +70,6 @@ impl ShareFsVolume {
                 }
             }
             Some(share_fs) => {
-                let readonly = m.options.iter().any(|opt| opt == "ro");
-
                 let share_fs_mount = share_fs.get_share_fs_mount();
                 let mounted_info_set = share_fs.mounted_info_set();
                 let mut mounted_info_set = mounted_info_set.lock().await;
@@ -226,6 +225,10 @@ impl Volume for ShareFsVolume {
 
         Ok(())
     }
+
+    fn get_device_id(&self) -> Result<Option<String>> {
+        Ok(None)
+    }
 }
 
 pub(crate) fn is_share_fs_volume(m: &oci::Mount) -> bool {

src/runtime-rs/crates/resource/src/volume/shm_volume.rs

@@ -104,6 +104,10 @@ impl Volume for ShmVolume {
         warn!(sl!(), "Cleaning up ShmVolume is still unimplemented.");
         Ok(())
     }
+
+    fn get_device_id(&self) -> Result<Option<String>> {
+        Ok(None)
+    }
 }
 
 pub(crate) fn is_shim_volume(m: &oci::Mount) -> bool {