devices: Add kernel cmdline, kernel, and initramfs to fw_cfg device

The kernel and initramfs are passed to the fw_cfg device as
file references. The cmdline is passed directly.

Signed-off-by: Alex Orozco <alexorozco@google.com>

Cargo.lock

@@ -531,6 +531,7 @@ dependencies = [
  "event_monitor",
  "hypervisor",
  "libc",
+ "linux-loader",
  "log",
  "num_enum",
  "pci",
@@ -1143,8 +1144,7 @@ dependencies = [
 [[package]]
 name = "linux-loader"
 version = "0.13.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "870c3814345f050991f99869417779f6062542bcf4ed81db7a1b926ad1306638"
+source = "git+https://github.com/rust-vmm/linux-loader?branch=main#d5f39c09d59c8f50d5313b78ce4de511b12d1848"
 dependencies = [
  "vm-memory",
 ]

Cargo.toml

@@ -107,7 +107,8 @@ members = [
 acpi_tables = { git = "https://github.com/rust-vmm/acpi_tables", branch = "main" }
 kvm-bindings = "0.12.0"
 kvm-ioctls = "0.22.0"
-linux-loader = "0.13.0"
+# TODO: update to 0.13.1+
+linux-loader = { git = "https://github.com/rust-vmm/linux-loader", branch = "main" }
 mshv-bindings = "0.5.2"
 mshv-ioctls = "0.5.2"
 seccompiler = "0.5.0"

devices/Cargo.toml

@@ -13,6 +13,11 @@ byteorder = { workspace = true }
 event_monitor = { path = "../event_monitor" }
 hypervisor = { path = "../hypervisor" }
 libc = { workspace = true }
+linux-loader = { workspace = true, features = [
+  "bzimage",
+  "elf",
+  "pe",
+], optional = true }
 log = { workspace = true }
 num_enum = "0.7.2"
 pci = { path = "../pci" }
@@ -38,6 +43,6 @@ arch = { path = "../arch" }
 
 [features]
 default = []
-fw_cfg = ["zerocopy"]
+fw_cfg = ["linux-loader", "zerocopy"]
 kvm = ["arch/kvm"]
 pvmemcontrol = []

devices/src/legacy/fw_cfg.rs

@@ -14,12 +14,17 @@
 use std::{
     fs::File,
     io::Result,
-    mem::size_of_val,
+    mem::{size_of, size_of_val},
     os::unix::fs::FileExt,
     sync::{Arc, Barrier},
 };
 
+#[cfg(target_arch = "x86_64")]
+use linux_loader::bootparam::boot_params;
+#[cfg(target_arch = "aarch64")]
+use linux_loader::loader::pe::arm64_image_header as boot_params;
 use vm_device::BusDevice;
+use vm_memory::ByteValued;
 use vmm_sys_util::sock_ctrl_msg::IntoIovec;
 use zerocopy::{FromBytes, IntoBytes};
 
@@ -50,6 +55,14 @@ pub const PORT_FW_CFG_WIDTH: u64 = 0x10;
 
 const FW_CFG_SIGNATURE: u16 = 0x00;
 const FW_CFG_ID: u16 = 0x01;
+const FW_CFG_KERNEL_SIZE: u16 = 0x08;
+const FW_CFG_INITRD_SIZE: u16 = 0x0b;
+const FW_CFG_KERNEL_DATA: u16 = 0x11;
+const FW_CFG_INITRD_DATA: u16 = 0x12;
+const FW_CFG_CMDLINE_SIZE: u16 = 0x14;
+const FW_CFG_CMDLINE_DATA: u16 = 0x15;
+const FW_CFG_SETUP_SIZE: u16 = 0x17;
+const FW_CFG_SETUP_DATA: u16 = 0x18;
 const FW_CFG_FILE_DIR: u16 = 0x19;
 const FW_CFG_KNOWN_ITEMS: usize = 0x20;
 
@@ -183,6 +196,46 @@ impl FwCfg {
         Ok(())
     }
 
+    pub fn add_kernel_data(&mut self, file: &File) -> Result<()> {
+        let mut buffer = vec![0u8; size_of::<boot_params>()];
+        file.read_exact_at(&mut buffer, 0)?;
+        let bp = boot_params::from_mut_slice(&mut buffer).unwrap();
+        #[cfg(target_arch = "x86_64")]
+        {
+            // must set to 4 for backwards compatibility
+            // https://docs.kernel.org/arch/x86/boot.html#the-real-mode-kernel-header
+            if bp.hdr.setup_sects == 0 {
+                bp.hdr.setup_sects = 4;
+            }
+            // wildcard boot loader type
+            bp.hdr.type_of_loader = 0xff;
+        }
+        #[cfg(target_arch = "aarch64")]
+        let kernel_start = bp.text_offset;
+        #[cfg(target_arch = "x86_64")]
+        let kernel_start = (bp.hdr.setup_sects as usize + 1) * 512;
+        self.known_items[FW_CFG_SETUP_SIZE as usize] = FwCfgContent::U32(buffer.len() as u32);
+        self.known_items[FW_CFG_SETUP_DATA as usize] = FwCfgContent::Bytes(buffer);
+        self.known_items[FW_CFG_KERNEL_SIZE as usize] =
+            FwCfgContent::U32(file.metadata()?.len() as u32 - kernel_start as u32);
+        self.known_items[FW_CFG_KERNEL_DATA as usize] =
+            FwCfgContent::File(kernel_start as u64, file.try_clone()?);
+        Ok(())
+    }
+
+    pub fn add_kernel_cmdline(&mut self, s: std::ffi::CString) {
+        let bytes = s.into_bytes_with_nul();
+        self.known_items[FW_CFG_CMDLINE_SIZE as usize] = FwCfgContent::U32(bytes.len() as u32);
+        self.known_items[FW_CFG_CMDLINE_DATA as usize] = FwCfgContent::Bytes(bytes);
+    }
+
+    pub fn add_initramfs_data(&mut self, file: &File) -> Result<()> {
+        let initramfs_size = file.metadata()?.len();
+        self.known_items[FW_CFG_INITRD_SIZE as usize] = FwCfgContent::U32(initramfs_size as _);
+        self.known_items[FW_CFG_INITRD_DATA as usize] = FwCfgContent::File(0, file.try_clone()?);
+        Ok(())
+    }
+
     fn read_content(content: &FwCfgContent, offset: u32, data: &mut [u8], size: u32) -> Option<u8> {
         let start = offset as usize;
         let end = start + size as usize;
@@ -278,6 +331,11 @@ impl BusDevice for FwCfg {
 
 #[cfg(test)]
 mod tests {
+    use std::ffi::CString;
+    use std::io::Write;
+
+    use vmm_sys_util::tempfile::TempFile;
+
     use super::*;
 
     #[cfg(target_arch = "x86_64")]
@@ -306,4 +364,51 @@ mod tests {
             }
         }
     }
+    #[test]
+    fn test_kernel_cmdline() {
+        let mut fw_cfg = FwCfg::new();
+
+        let cmdline = *b"cmdline\0";
+
+        fw_cfg.add_kernel_cmdline(CString::from_vec_with_nul(cmdline.to_vec()).unwrap());
+
+        let mut data = vec![0u8];
+
+        let mut cmdline_iter = cmdline.into_iter();
+        fw_cfg.write(0, SELECTOR_OFFSET, &[FW_CFG_CMDLINE_DATA as u8, 0]);
+        loop {
+            if let Some(char) = cmdline_iter.next() {
+                fw_cfg.read(0, DATA_OFFSET, &mut data);
+                assert_eq!(data[0], char);
+            } else {
+                return;
+            }
+        }
+    }
+
+    #[test]
+    fn test_initram_fs() {
+        let mut fw_cfg = FwCfg::new();
+
+        let temp = TempFile::new().unwrap();
+        let mut temp_file = temp.as_file();
+
+        let initram_content = b"this is the initramfs";
+        let written = temp_file.write(initram_content);
+        assert_eq!(written.unwrap(), 21);
+        let _ = fw_cfg.add_initramfs_data(temp_file);
+
+        let mut data = vec![0u8];
+
+        let mut initram_iter = (*initram_content).into_iter();
+        fw_cfg.write(0, SELECTOR_OFFSET, &[FW_CFG_INITRD_DATA as u8, 0]);
+        loop {
+            if let Some(char) = initram_iter.next() {
+                fw_cfg.read(0, DATA_OFFSET, &mut data);
+                assert_eq!(data[0], char);
+            } else {
+                return;
+            }
+        }
+    }
 }

fuzz/Cargo.toml

@@ -22,7 +22,12 @@ epoll = "4.3.3"
 hypervisor = { path = "../hypervisor", features = ["mshv_emulator"] }
 libc = "0.2.155"
 libfuzzer-sys = "0.4.7"
-linux-loader = { version = "0.13.0", features = ["bzimage", "elf", "pe"] }
+# TODO: update to 0.13.1+
+linux-loader = { git = "https://github.com/rust-vmm/linux-loader", branch = "main", features = [
+  "bzimage",
+  "elf",
+  "pe",
+] }
 micro_http = { git = "https://github.com/firecracker-microvm/micro-http", branch = "main" }
 mshv-bindings = "0.5.2"
 net_util = { path = "../net_util" }

vmm/src/vm.rs

@@ -339,6 +339,18 @@ pub enum Error {
 
     #[error("Error locking disk images: Another instance likely holds a lock")]
     LockingError(#[source] DeviceManagerError),
+
+    #[cfg(feature = "fw_cfg")]
+    #[error("Fw Cfg missing kernel")]
+    MissingFwCfgKernelFile(#[source] io::Error),
+
+    #[cfg(feature = "fw_cfg")]
+    #[error("Fw Cfg missing initramfs")]
+    MissingFwCfgInitramfs(#[source] io::Error),
+
+    #[cfg(feature = "fw_cfg")]
+    #[error("Fw Cfg missing kernel cmdline")]
+    MissingFwCfgCmdline,
 }
 pub type Result<T> = result::Result<T, Error>;
 
@@ -784,6 +796,72 @@ impl Vm {
         })
     }
 
+    #[cfg(feature = "fw_cfg")]
+    fn populate_fw_cfg(
+        device_manager: &Arc<Mutex<DeviceManager>>,
+        config: &Arc<Mutex<VmConfig>>,
+    ) -> Result<()> {
+        let kernel = config
+            .lock()
+            .unwrap()
+            .payload
+            .as_ref()
+            .map(|p| p.kernel.as_ref().map(File::open))
+            .unwrap_or_default()
+            .transpose()
+            .map_err(Error::MissingFwCfgKernelFile)?;
+        if let Some(kernel_file) = kernel {
+            device_manager
+                .lock()
+                .unwrap()
+                .fw_cfg()
+                .expect("fw_cfg device must be present")
+                .lock()
+                .unwrap()
+                .add_kernel_data(&kernel_file)
+                .map_err(Error::MissingFwCfgKernelFile)?
+        }
+        let cmdline = Vm::generate_cmdline(
+            config.lock().unwrap().payload.as_ref().unwrap(),
+            #[cfg(target_arch = "aarch64")]
+            device_manager,
+        )
+        .map_err(|_| Error::MissingFwCfgCmdline)?
+        .as_cstring()
+        .map_err(|_| Error::MissingFwCfgCmdline)?;
+        device_manager
+            .lock()
+            .unwrap()
+            .fw_cfg()
+            .expect("fw_cfg device must be present")
+            .lock()
+            .unwrap()
+            .add_kernel_cmdline(cmdline);
+        let initramfs = config
+            .lock()
+            .unwrap()
+            .payload
+            .as_ref()
+            .map(|p| p.initramfs.as_ref().map(File::open))
+            .unwrap_or_default()
+            .transpose()
+            .map_err(Error::MissingFwCfgInitramfs)?;
+        // We measure the initramfs when running Oak Containers in SNP mode (initramfs = Stage1)
+        // o/w use Stage0 to launch cloud disk images
+        if let Some(initramfs_file) = initramfs {
+            device_manager
+                .lock()
+                .unwrap()
+                .fw_cfg()
+                .expect("fw_cfg device must be present")
+                .lock()
+                .unwrap()
+                .add_initramfs_data(&initramfs_file)
+                .map_err(Error::MissingFwCfgInitramfs)?;
+        }
+        Ok(())
+    }
+
     fn create_numa_nodes(
         configs: Option<Vec<NumaConfig>>,
         memory_manager: &Arc<Mutex<MemoryManager>>,
@@ -2272,6 +2350,9 @@ impl Vm {
         };
         current_state.valid_transition(new_state)?;
 
+        #[cfg(feature = "fw_cfg")]
+        Self::populate_fw_cfg(&self.device_manager, &self.config)?;
+
         // Do earlier to parallelise with loading kernel
         #[cfg(target_arch = "x86_64")]
         cfg_if::cfg_if! {