Merge pull request #12807 from PiotrProkop/blk-sector-rust

runtime-rs: allow specifying logical/physical sector size for block devices

src/libs/kata-types/src/annotations/mod.rs

@@ -284,6 +284,20 @@ pub const KATA_ANNO_CFG_HYPERVISOR_DEFAULT_GPUS: &str =
 pub const KATA_ANNO_CFG_HYPERVISOR_DEFAULT_GPU_MODEL: &str =
     "io.katacontainers.config.hypervisor.default_gpu_model";
 
+/// A sandbox annotation that specifies the logical sector size reported by block devices to the
+/// guest, in bytes. Common values are 512 and 4096. Set to 0 to use the hypervisor default.
+/// NOTE: the annotation key uses "blk_logical_sector_size" rather than
+/// "block_device_logical_sector_size" because Kubernetes enforces a 63-character limit on
+/// annotation name segments.
+pub const KATA_ANNO_CFG_HYPERVISOR_BLK_LOGICAL_SECTOR_SIZE: &str =
+    "io.katacontainers.config.hypervisor.blk_logical_sector_size";
+/// A sandbox annotation that specifies the physical sector size reported by block devices to the
+/// guest, in bytes. Common values are 512 and 4096. Set to 0 to use the hypervisor default.
+/// NOTE: the annotation key uses "blk_physical_sector_size" rather than
+/// "block_device_physical_sector_size" because Kubernetes enforces a 63-character limit on
+/// annotation name segments.
+pub const KATA_ANNO_CFG_HYPERVISOR_BLK_PHYSICAL_SECTOR_SIZE: &str =
+    "io.katacontainers.config.hypervisor.blk_physical_sector_size";
 /// Block device specific annotation for num_queues
 pub const KATA_ANNO_CFG_HYPERVISOR_BLOCK_DEV_NUM_QUEUES: &str =
     "io.katacontainers.config.hypervisor.block_device_num_queues";
@@ -973,6 +987,48 @@ impl Annotation {
                             hv.shared_fs.virtio_fs_extra_args.push(arg.to_string());
                         }
                     }
+                    KATA_ANNO_CFG_HYPERVISOR_BLK_LOGICAL_SECTOR_SIZE => {
+                        match self.get_value::<u32>(key) {
+                            Ok(v) => {
+                                let size = v.unwrap_or_default();
+                                if let Err(e) =
+                                    crate::config::hypervisor::validate_block_device_sector_size(
+                                        size,
+                                    )
+                                {
+                                    return Err(io::Error::new(
+                                        io::ErrorKind::InvalidData,
+                                        e.to_string(),
+                                    ));
+                                }
+                                hv.blockdev_info.block_device_logical_sector_size = size;
+                            }
+                            Err(_e) => {
+                                return Err(u32_err);
+                            }
+                        }
+                    }
+                    KATA_ANNO_CFG_HYPERVISOR_BLK_PHYSICAL_SECTOR_SIZE => {
+                        match self.get_value::<u32>(key) {
+                            Ok(v) => {
+                                let size = v.unwrap_or_default();
+                                if let Err(e) =
+                                    crate::config::hypervisor::validate_block_device_sector_size(
+                                        size,
+                                    )
+                                {
+                                    return Err(io::Error::new(
+                                        io::ErrorKind::InvalidData,
+                                        e.to_string(),
+                                    ));
+                                }
+                                hv.blockdev_info.block_device_physical_sector_size = size;
+                            }
+                            Err(_e) => {
+                                return Err(u32_err);
+                            }
+                        }
+                    }
                     KATA_ANNO_CFG_HYPERVISOR_BLOCK_DEV_NUM_QUEUES => {
                         match self.get_value::<usize>(key) {
                             Ok(v) => {
@@ -1123,6 +1179,18 @@ impl Annotation {
             }
         }
 
+        // Validate cross-field constraint: logical sector size must not exceed physical.
+        // Individual sizes are validated inside the loop, but the cross-field check must
+        // run after both annotations have been applied.
+        let logical = hv.blockdev_info.block_device_logical_sector_size;
+        let physical = hv.blockdev_info.block_device_physical_sector_size;
+        if logical != 0 && physical != 0 && logical > physical {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                format!("invalid sector sizes: logical ({logical}) must not be larger than physical ({physical})"),
+            ));
+        }
+
         config.adjust_config()?;
 
         Ok(())

src/libs/kata-types/src/config/hypervisor/mod.rs

@@ -271,6 +271,18 @@ pub struct BlockDeviceInfo {
     #[serde(default)]
     pub block_device_cache_noflush: bool,
 
+    /// Specifies the logical sector size, in bytes, reported by block devices to the guest.
+    /// Common values are 512 and 4096. Set to 0 to use the hypervisor default.
+    /// Must be 0 or a power of 2 between 512 and 65536.
+    #[serde(default)]
+    pub block_device_logical_sector_size: u32,
+
+    /// Specifies the physical sector size, in bytes, reported by block devices to the guest.
+    /// Common values are 512 and 4096. Set to 0 to use the hypervisor default.
+    /// Must be 0 or a power of 2 between 512 and 65536.
+    #[serde(default)]
+    pub block_device_physical_sector_size: u32,
+
     /// If false and nvdimm is supported, use nvdimm device to plug guest image.
     #[serde(default)]
     pub disable_image_nvdimm: bool,
@@ -400,6 +412,16 @@ impl BlockDeviceInfo {
             "Invalid vhost-user-store-path {}: {}"
         )?;
 
+        validate_block_device_sector_size(self.block_device_logical_sector_size)?;
+        validate_block_device_sector_size(self.block_device_physical_sector_size)?;
+        let logical = self.block_device_logical_sector_size;
+        let physical = self.block_device_physical_sector_size;
+        if logical != 0 && physical != 0 && logical > physical {
+            return Err(std::io::Error::other(format!(
+                "invalid sector sizes: logical ({logical}) must not be larger than physical ({physical})"
+            )));
+        }
+
         Ok(())
     }
 
@@ -409,6 +431,19 @@ impl BlockDeviceInfo {
     }
 }
 
+/// Validate that a block device sector size is 0 or a power of 2 in [512, 65536].
+pub fn validate_block_device_sector_size(size: u32) -> Result<()> {
+    if size == 0 {
+        return Ok(());
+    }
+    if !(512..=65536).contains(&size) || (size & (size - 1)) != 0 {
+        return Err(std::io::Error::other(format!(
+            "invalid sector size {size}: must be 0 or a power of 2 between 512 and 65536"
+        )));
+    }
+    Ok(())
+}
+
 /// Guest kernel boot information.
 #[derive(Clone, Debug, Default, Deserialize, Serialize)]
 pub struct BootInfo {
@@ -2072,4 +2107,83 @@ mod tests {
             expected_error_msg
         );
     }
+
+    #[test]
+    fn test_validate_block_device_sector_size_valid() {
+        for size in [0, 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536] {
+            assert!(
+                validate_block_device_sector_size(size).is_ok(),
+                "expected size {} to be accepted",
+                size
+            );
+        }
+    }
+
+    #[test]
+    fn test_validate_block_device_sector_size_not_power_of_two() {
+        for size in [3, 100, 1000, 3000, 5000] {
+            assert!(
+                validate_block_device_sector_size(size).is_err(),
+                "expected non-power-of-2 size {} to be rejected",
+                size
+            );
+        }
+    }
+
+    #[test]
+    fn test_validate_block_device_sector_size_below_minimum() {
+        for size in [1, 256] {
+            assert!(
+                validate_block_device_sector_size(size).is_err(),
+                "expected below-minimum size {} to be rejected",
+                size
+            );
+        }
+    }
+
+    #[test]
+    fn test_validate_block_device_sector_size_above_maximum() {
+        for size in [131072, 1048576] {
+            assert!(
+                validate_block_device_sector_size(size).is_err(),
+                "expected above-maximum size {} to be rejected",
+                size
+            );
+        }
+    }
+
+    fn blockdev_info_with_sectors(logical: u32, physical: u32) -> BlockDeviceInfo {
+        BlockDeviceInfo {
+            block_device_driver: VIRTIO_BLK_PCI.to_string(),
+            block_device_logical_sector_size: logical,
+            block_device_physical_sector_size: physical,
+            ..Default::default()
+        }
+    }
+
+    #[test]
+    fn test_validate_block_device_sector_sizes_valid() {
+        assert!(blockdev_info_with_sectors(0, 0).validate().is_ok());
+        assert!(blockdev_info_with_sectors(512, 0).validate().is_ok());
+        assert!(blockdev_info_with_sectors(0, 4096).validate().is_ok());
+        assert!(blockdev_info_with_sectors(512, 4096).validate().is_ok());
+        assert!(blockdev_info_with_sectors(4096, 4096).validate().is_ok());
+        assert!(blockdev_info_with_sectors(512, 512).validate().is_ok());
+    }
+
+    #[test]
+    fn test_validate_block_device_sector_sizes_logical_exceeds_physical() {
+        assert!(
+            blockdev_info_with_sectors(4096, 512).validate().is_err(),
+            "logical > physical should be rejected"
+        );
+        assert!(
+            blockdev_info_with_sectors(4096, 1024).validate().is_err(),
+            "logical > physical should be rejected"
+        );
+        assert!(
+            blockdev_info_with_sectors(65536, 512).validate().is_err(),
+            "logical > physical should be rejected"
+        );
+    }
 }

src/libs/kata-types/tests/test_config.rs

@@ -7,7 +7,9 @@ mod tests {
     use kata_types::annotations::{
         Annotation, KATA_ANNO_CFG_AGENT_CONTAINER_PIPE_SIZE, KATA_ANNO_CFG_AGENT_TRACE,
         KATA_ANNO_CFG_DISABLE_GUEST_SECCOMP, KATA_ANNO_CFG_ENABLE_PPROF,
-        KATA_ANNO_CFG_EXPERIMENTAL, KATA_ANNO_CFG_HYPERVISOR_BLOCK_DEV_CACHE_NOFLUSH,
+        KATA_ANNO_CFG_EXPERIMENTAL, KATA_ANNO_CFG_HYPERVISOR_BLK_LOGICAL_SECTOR_SIZE,
+        KATA_ANNO_CFG_HYPERVISOR_BLK_PHYSICAL_SECTOR_SIZE,
+        KATA_ANNO_CFG_HYPERVISOR_BLOCK_DEV_CACHE_NOFLUSH,
         KATA_ANNO_CFG_HYPERVISOR_BLOCK_DEV_DRIVER, KATA_ANNO_CFG_HYPERVISOR_DEFAULT_MEMORY,
         KATA_ANNO_CFG_HYPERVISOR_DEFAULT_VCPUS, KATA_ANNO_CFG_HYPERVISOR_ENABLE_GUEST_SWAP,
         KATA_ANNO_CFG_HYPERVISOR_ENABLE_HUGEPAGES, KATA_ANNO_CFG_HYPERVISOR_ENABLE_IO_THREADS,
@@ -479,4 +481,111 @@ mod tests {
         let mut config = TomlConfig::load(content).unwrap();
         assert!(anno.update_config_by_annotation(&mut config).is_err());
     }
+
+    #[test]
+    fn test_block_device_sector_size_annotations_valid() {
+        let content = include_str!("texture/configuration-anno-0.toml");
+
+        let qemu = QemuConfig::new();
+        qemu.register();
+
+        // Valid: 512 logical, 4096 physical
+        let config = TomlConfig::load(content).unwrap();
+        KataConfig::set_active_config(Some(config), "qemu", "agent0");
+
+        let mut anno_hash = HashMap::new();
+        anno_hash.insert(
+            KATA_ANNO_CFG_HYPERVISOR_BLK_LOGICAL_SECTOR_SIZE.to_string(),
+            "512".to_string(),
+        );
+        anno_hash.insert(
+            KATA_ANNO_CFG_HYPERVISOR_BLK_PHYSICAL_SECTOR_SIZE.to_string(),
+            "4096".to_string(),
+        );
+        let anno = Annotation::new(anno_hash);
+        let mut config = TomlConfig::load(content).unwrap();
+        assert!(anno.update_config_by_annotation(&mut config).is_ok());
+        if let Some(hv) = config.hypervisor.get("qemu") {
+            assert_eq!(hv.blockdev_info.block_device_logical_sector_size, 512);
+            assert_eq!(hv.blockdev_info.block_device_physical_sector_size, 4096);
+        }
+
+        // Valid: 0 means hypervisor default
+        let mut anno_hash = HashMap::new();
+        anno_hash.insert(
+            KATA_ANNO_CFG_HYPERVISOR_BLK_LOGICAL_SECTOR_SIZE.to_string(),
+            "0".to_string(),
+        );
+        anno_hash.insert(
+            KATA_ANNO_CFG_HYPERVISOR_BLK_PHYSICAL_SECTOR_SIZE.to_string(),
+            "0".to_string(),
+        );
+        let anno = Annotation::new(anno_hash);
+        let mut config = TomlConfig::load(content).unwrap();
+        assert!(anno.update_config_by_annotation(&mut config).is_ok());
+        if let Some(hv) = config.hypervisor.get("qemu") {
+            assert_eq!(hv.blockdev_info.block_device_logical_sector_size, 0);
+            assert_eq!(hv.blockdev_info.block_device_physical_sector_size, 0);
+        }
+    }
+
+    #[test]
+    fn test_block_device_sector_size_annotation_invalid_not_power_of_two() {
+        let content = include_str!("texture/configuration-anno-0.toml");
+
+        let qemu = QemuConfig::new();
+        qemu.register();
+
+        let config = TomlConfig::load(content).unwrap();
+        KataConfig::set_active_config(Some(config), "qemu", "agent0");
+
+        let mut anno_hash = HashMap::new();
+        anno_hash.insert(
+            KATA_ANNO_CFG_HYPERVISOR_BLK_LOGICAL_SECTOR_SIZE.to_string(),
+            "1000".to_string(),
+        );
+        let anno = Annotation::new(anno_hash);
+        let mut config = TomlConfig::load(content).unwrap();
+        assert!(anno.update_config_by_annotation(&mut config).is_err());
+    }
+
+    #[test]
+    fn test_block_device_sector_size_annotation_invalid_below_minimum() {
+        let content = include_str!("texture/configuration-anno-0.toml");
+
+        let qemu = QemuConfig::new();
+        qemu.register();
+
+        let config = TomlConfig::load(content).unwrap();
+        KataConfig::set_active_config(Some(config), "qemu", "agent0");
+
+        let mut anno_hash = HashMap::new();
+        anno_hash.insert(
+            KATA_ANNO_CFG_HYPERVISOR_BLK_PHYSICAL_SECTOR_SIZE.to_string(),
+            "256".to_string(),
+        );
+        let anno = Annotation::new(anno_hash);
+        let mut config = TomlConfig::load(content).unwrap();
+        assert!(anno.update_config_by_annotation(&mut config).is_err());
+    }
+
+    #[test]
+    fn test_block_device_sector_size_annotation_invalid_above_maximum() {
+        let content = include_str!("texture/configuration-anno-0.toml");
+
+        let qemu = QemuConfig::new();
+        qemu.register();
+
+        let config = TomlConfig::load(content).unwrap();
+        KataConfig::set_active_config(Some(config), "qemu", "agent0");
+
+        let mut anno_hash = HashMap::new();
+        anno_hash.insert(
+            KATA_ANNO_CFG_HYPERVISOR_BLK_LOGICAL_SECTOR_SIZE.to_string(),
+            "131072".to_string(),
+        );
+        let anno = Annotation::new(anno_hash);
+        let mut config = TomlConfig::load(content).unwrap();
+        assert!(anno.update_config_by_annotation(&mut config).is_err());
+    }
 }

src/libs/kata-types/tests/texture/configuration-anno-0.toml

@@ -19,7 +19,7 @@ default_maxvcpus = 64
 machine_type = "q35"
 confidential_guest = true
 rootless = true
-enable_annotations = ["shared_fs","path", "ctlpath","jailer_path","enable_iothreads","default_memory","memory_slots","enable_mem_prealloc","enable_hugepages","file_mem_backend","enable_virtio_mem","enable_guest_swap","default_vcpus","virtio_fs_extra_args","block_device_driver","vhost_user_store_path","kernel","guest_hook_path","block_device_cache_noflush","virtio_fs_daemon"]
+enable_annotations = ["shared_fs","path", "ctlpath","jailer_path","enable_iothreads","default_memory","memory_slots","enable_mem_prealloc","enable_hugepages","file_mem_backend","enable_virtio_mem","enable_guest_swap","default_vcpus","virtio_fs_extra_args","block_device_driver","vhost_user_store_path","kernel","guest_hook_path","block_device_cache_noflush","virtio_fs_daemon","blk_logical_sector_size","blk_physical_sector_size"]
 machine_accelerators="noapic"
 default_bridges = 2
 default_memory = 128

src/runtime-rs/config/configuration-qemu-coco-dev-runtime-rs.toml.in

@@ -255,6 +255,16 @@ block_device_cache_direct = false
 # Default false
 block_device_cache_noflush = false
 
+# Specifies the logical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_logical_sector_size = 0
+
+# Specifies the physical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_physical_sector_size = 0
+
 # Enable iothreads (data-plane) to be used. This causes IO to be
 # handled in a separate IO thread. This is currently only implemented
 # for SCSI.

src/runtime-rs/config/configuration-qemu-runtime-rs.toml.in

@@ -245,6 +245,16 @@ block_device_cache_direct = false
 # Default false
 block_device_cache_noflush = false
 
+# Specifies the logical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_logical_sector_size = 0
+
+# Specifies the physical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_physical_sector_size = 0
+
 # Enable iothreads (data-plane) to be used. This causes IO to be
 # handled in a separate IO thread. This is currently only implemented
 # for SCSI.

src/runtime-rs/config/configuration-qemu-se-runtime-rs.toml.in

@@ -244,6 +244,16 @@ block_device_cache_direct = false
 # Default false
 block_device_cache_noflush = false
 
+# Specifies the logical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_logical_sector_size = 0
+
+# Specifies the physical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_physical_sector_size = 0
+
 # Enable iothreads (data-plane) to be used. This causes IO to be
 # handled in a separate IO thread. This is currently only implemented
 # for SCSI.

src/runtime-rs/config/configuration-qemu-snp-runtime-rs.toml.in

@@ -281,6 +281,16 @@ block_device_cache_direct = false
 # Default false
 block_device_cache_noflush = false
 
+# Specifies the logical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_logical_sector_size = 0
+
+# Specifies the physical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_physical_sector_size = 0
+
 # Enable iothreads (data-plane) to be used. This causes IO to be
 # handled in a separate IO thread. This is currently only implemented
 # for SCSI.

src/runtime-rs/config/configuration-qemu-tdx-runtime-rs.toml.in

@@ -256,6 +256,16 @@ block_device_cache_direct = false
 # Default false
 block_device_cache_noflush = false
 
+# Specifies the logical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_logical_sector_size = 0
+
+# Specifies the physical sector size, in bytes, reported by block devices to the guest.
+# Common values are 512 and 4096. Set to 0 to use the QEMU/hypervisor default.
+# Default 0
+block_device_physical_sector_size = 0
+
 # Enable iothreads (data-plane) to be used. This causes IO to be
 # handled in a separate IO thread. This is currently implemented
 # for virtio-scsi and virtio-blk.

src/runtime-rs/crates/hypervisor/src/device/driver/virtio_blk.rs

@@ -112,6 +112,12 @@ pub struct BlockConfig {
 
     /// block device multi-queue
     pub num_queues: usize,
+
+    /// Logical sector size in bytes reported to the guest. 0 means use hypervisor default.
+    pub logical_sector_size: u32,
+
+    /// Physical sector size in bytes reported to the guest. 0 means use hypervisor default.
+    pub physical_sector_size: u32,
 }
 
 #[derive(Debug, Clone, Default)]

src/runtime-rs/crates/hypervisor/src/qemu/inner.rs

@@ -866,6 +866,8 @@ impl QemuInner {
                         ),
                         block_device.config.is_readonly,
                         block_device.config.no_drop,
+                        block_device.config.logical_sector_size,
+                        block_device.config.physical_sector_size,
                     )
                     .context("hotplug block device")?;
 

src/runtime-rs/crates/hypervisor/src/qemu/qmp.rs

@@ -642,6 +642,8 @@ impl Qmp {
         is_direct: Option<bool>,
         is_readonly: bool,
         no_drop: bool,
+        logical_block_size: u32,
+        physical_block_size: u32,
     ) -> Result<(Option<PciPath>, Option<String>)> {
         // `blockdev-add`
         let node_name = format!("drive-{index}");
@@ -719,6 +721,13 @@ impl Qmp {
         let mut blkdev_add_args = Dictionary::new();
         blkdev_add_args.insert("drive".to_owned(), node_name.clone().into());
 
+        if logical_block_size > 0 {
+            blkdev_add_args.insert("logical_block_size".to_owned(), logical_block_size.into());
+        }
+        if physical_block_size > 0 {
+            blkdev_add_args.insert("physical_block_size".to_owned(), physical_block_size.into());
+        }
+
         if block_driver == VIRTIO_SCSI {
             // Helper closure to decode a flattened u16 SCSI index into an (ID, LUN) pair.
             let get_scsi_id_lun = |index_u16: u16| -> Result<(u8, u8)> {

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

@@ -422,6 +422,8 @@ impl ResourceManagerInner {
                         blkdev_aio: BlockDeviceAio::new(&blkdev_info.block_device_aio),
                         num_queues: blkdev_info.num_queues,
                         queue_size: blkdev_info.queue_size,
+                        logical_sector_size: blkdev_info.block_device_logical_sector_size,
+                        physical_sector_size: blkdev_info.block_device_physical_sector_size,
                         ..Default::default()
                     });
 

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

@@ -49,6 +49,8 @@ impl BlockVolume {
             blkdev_aio: BlockDeviceAio::new(&blkdev_info.block_device_aio),
             num_queues: blkdev_info.num_queues,
             queue_size: blkdev_info.queue_size,
+            logical_sector_size: blkdev_info.block_device_logical_sector_size,
+            physical_sector_size: blkdev_info.block_device_physical_sector_size,
             ..Default::default()
         };
 

src/runtime-rs/crates/resource/src/volume/direct_volumes/rawblock_volume.rs

@@ -64,6 +64,8 @@ impl RawblockVolume {
             blkdev_aio: BlockDeviceAio::new(&blkdev_info.block_device_aio),
             num_queues: blkdev_info.num_queues,
             queue_size: blkdev_info.queue_size,
+            logical_sector_size: blkdev_info.block_device_logical_sector_size,
+            physical_sector_size: blkdev_info.block_device_physical_sector_size,
             ..Default::default()
         };