코드정리, kmsg 기반으로 로그 전송하도록 처리, allocator 메모리 재활용하도록 처리, init 찾을 수 있도록 처리

Dockerfile

@@ -10,7 +10,7 @@ LABEL org.opencontainers.image.authors="Sangbum Kim <sangbumkim@amuz.es>"
 WORKDIR /app
 COPY . /app
 
-ENV RUSTFLAGS='-C link-arg=-s -C link-arg=-fuse-ld=lld'
+ENV RUSTFLAGS='-C panic=abort -C link-arg=-s -C link-arg=-fuse-ld=lld'
 
 RUN set -x && \
     apk add --no-cache \
@@ -22,6 +22,7 @@ RUN set -x && \
 RUN set -x && \
     cargo build --release && \
     ls -alh target/release/init-wrapper
+    # ldd target/release/init-wrapper && \
     # && \
     # ldd target/release/init-wrapper
 

README.md

@@ -16,6 +16,7 @@ mount -t overlay -o rw,relatime,lowerdir=/run/overlay/lower,upperdir=/run/overla
 mkdir -p /run/overlay/merged/oldroot
 pivot_root /run/overlay/merged /run/overlay/merged/oldroot
 cd /
+mount  --move /oldroot/dev /dev || true
 mount  --move /oldroot/run /run
 umount -l /oldroot
 rmdir /oldroot

src/kmsg.rs

@@ -0,0 +1,59 @@
+#![allow(unused_imports)]
+
+use alloc::format;
+use core::ffi::CStr;
+use libc::{c_char, c_int, open, FILE, O_APPEND, O_CLOEXEC, O_SYNC, O_WRONLY, STDERR_FILENO};
+use libc_print::__LibCWriter;
+pub(crate) static mut LOG_FD: c_int = STDERR_FILENO;
+pub(crate) static mut LOG_FD_KMSG: bool = false;
+
+#[inline]
+pub(crate) fn init_kmsg() {
+    const DEV_KMSG_PATH: *const c_char =
+        unsafe { CStr::from_bytes_with_nul_unchecked(b"/dev/kmsg\0").as_ptr() };
+    const LOG_FLAGS: c_int = O_APPEND | O_WRONLY | O_CLOEXEC | O_SYNC;
+
+    let fd = unsafe { open(DEV_KMSG_PATH, LOG_FLAGS) };
+    match fd {
+        -1 => (),
+        fd => unsafe {
+            crate::kmsg::LOG_FD = fd;
+            LOG_FD_KMSG = true;
+        },
+    }
+}
+
+#[macro_export]
+macro_rules! kprintln {
+    () => { $crate::kprintln!("") };
+    ($fmt: tt) => {
+        {
+            #[allow(unused_must_use)]
+            unsafe {
+                let mut stm = libc_print::__LibCWriter::new($crate::kmsg::LOG_FD);
+                stm.write_str(concat!("init-wrapper: ",$fmt));
+                if (!$crate::kmsg::LOG_FD_KMSG) {
+                    stm.write_nl();
+                }
+            }
+        }
+   };
+    ($fmt: tt, $($arg:tt)*) =>{
+        {
+
+            #[allow(unused_must_use)]
+            unsafe {
+                let mut stm = libc_print::__LibCWriter::new($crate::kmsg::LOG_FD);
+                if ($crate::kmsg::LOG_FD_KMSG) {
+                    let buf=alloc::format!(concat!("init-wrapper: ",$fmt),$($arg)*);
+                    for line in buf.lines(){
+                        stm.write_str(&line);
+                    }
+                }else{
+                    stm.write_fmt(format_args!(concat!("init-wrapper: ",$fmt),$($arg)*));
+                    stm.write_nl();
+                }
+            }
+        }
+    };
+}

src/link.rs

@@ -1,6 +1,6 @@
 #![allow(non_camel_case_types, dead_code)]
 
-use libc::FILE;
+use libc::{exit, FILE};
 
 #[link(name = "c")]
 extern "C" {
@@ -9,9 +9,16 @@ extern "C" {
     static stderr: *mut FILE;
 }
 
+#[no_mangle]
+extern "C" fn rust_eh_personality() {}
+#[no_mangle]
+unsafe extern "C" fn _Unwind_Resume() -> ! {
+    exit(2)
+}
+
 #[panic_handler]
 #[inline(never)]
 #[cfg(not(test))]
 unsafe fn panic_handler(_: &core::panic::PanicInfo) -> ! {
-    libc::exit(2)
+    exit(2)
 }

src/main.rs

@@ -2,167 +2,30 @@
 #![no_main]
 
 extern crate alloc;
+mod kmsg;
 mod link;
 mod mem;
+mod tm;
+mod unix;
+use errno::Errno;
 
-use alloc::ffi::CString;
-use core::ffi::CStr;
-use core::ops::Sub;
-use core::ptr::null;
-use core::time::Duration;
-use errno::{errno, Errno};
+use kmsg::init_kmsg;
 use libc::{
-    c_int, c_void, chdir, clock_gettime, execv, mkdir, mode_t, mount, rmdir, syscall, timespec,
-    umount2, SYS_pivot_root, CLOCK_BOOTTIME, EXIT_FAILURE, EXIT_SUCCESS, MNT_DETACH, MS_BIND,
-    MS_LAZYTIME, MS_MOVE, MS_NODEV, MS_NOSUID, MS_PRIVATE, MS_RDONLY, MS_REC, MS_RELATIME,
+    c_int, EACCES, EINVAL, ENOENT, ENOEXEC, ENOTDIR, EXIT_FAILURE, EXIT_SUCCESS, MNT_DETACH,
+    MS_BIND, MS_LAZYTIME, MS_MOVE, MS_NODEV, MS_NOSUID, MS_PRIVATE, MS_RDONLY, MS_REC, MS_RELATIME,
 };
-use libc_print::std_name::eprintln;
 
-struct Timespec {
-    ts: timespec,
-}
-
-type SystemResult = Result<(), Errno>;
-
-impl Sub<Timespec> for Timespec {
-    type Output = Duration;
-
-    fn sub(self, other: Timespec) -> Duration {
-        let sec = self.ts.tv_sec - other.ts.tv_sec;
-        let nsec = self.ts.tv_nsec - other.ts.tv_nsec;
-        if nsec < 0 {
-            Duration::from_secs(sec as u64) - Duration::from_nanos(nsec.unsigned_abs())
-        } else {
-            Duration::from_secs(sec as u64) + Duration::from_nanos(nsec.unsigned_abs())
-        }
-    }
-}
-
-impl From<timespec> for Timespec {
-    fn from(item: timespec) -> Self {
-        Timespec { ts: item }
-    }
-}
-
-impl Into<timespec> for Timespec {
-    fn into(self) -> timespec {
-        self.ts
-    }
-}
-
-fn do_mount(
-    source: Option<&str>,
-    target: Option<&str>,
-    fs: Option<&str>,
-    flags: u64,
-    opt: Option<&str>,
-) -> SystemResult {
-    // has ownership
-    let raw_src = source.map(|v| CString::new(v).ok()).flatten();
-    let raw_tgt = target.map(|v| CString::new(v).ok()).flatten();
-    let raw_fs = fs.map(|v| CString::new(v).ok()).flatten();
-    let raw_fs_opt = opt.map(|v| CString::new(v).ok()).flatten();
-
-    unsafe {
-        if mount(
-            raw_src.as_ref().map(|v| v.as_ptr()).unwrap_or_else(null),
-            raw_tgt.as_ref().map(|v| v.as_ptr()).unwrap_or_else(null),
-            raw_fs.as_ref().map(|v| v.as_ptr()).unwrap_or_else(null),
-            flags,
-            raw_fs_opt.as_ref().map(|v| v.as_ptr()).unwrap_or_else(null) as *const c_void,
-        ) == -1
-        {
-            Err(errno())
-        } else {
-            Ok(())
-        }
-    }
-}
-fn do_umount(path: &str, flags: i32) -> SystemResult {
-    // has ownership
-    let raw_path = CString::new(path).unwrap();
-    unsafe {
-        if umount2(raw_path.as_ref().as_ptr(), flags) == -1 {
-            Err(errno())
-        } else {
-            Ok(())
-        }
-    }
-}
-
-fn do_mkdir(path: &str, mode: mode_t) -> SystemResult {
-    // has ownership
-    let raw_path = CString::new(path).unwrap();
-    unsafe {
-        if mkdir(raw_path.as_ref().as_ptr(), mode) == -1 {
-            Err(errno())
-        } else {
-            Ok(())
-        }
-    }
-}
-fn do_rmdir(path: &str) -> SystemResult {
-    // has ownership
-    let raw_path = CString::new(path).unwrap();
-    unsafe {
-        if rmdir(raw_path.as_ref().as_ptr()) == -1 {
-            Err(errno())
-        } else {
-            Ok(())
-        }
-    }
-}
-fn do_chdir(path: &str) -> SystemResult {
-    // has ownership
-    let raw_path = CString::new(path).unwrap();
-    unsafe {
-        if chdir(raw_path.as_ref().as_ptr()) == -1 {
-            Err(errno())
-        } else {
-            Ok(())
-        }
-    }
-}
-
-fn pivot_root(new_root: &str, put_old: &str) -> SystemResult {
-    let raw_new_root = CString::new(new_root).unwrap();
-    let raw_put_old = CString::new(put_old).unwrap();
-    unsafe {
-        if syscall(
-            SYS_pivot_root,
-            raw_new_root.as_ref().as_ptr(),
-            raw_put_old.as_ref().as_ptr(),
-        ) == -1
-        {
-            Err(errno())
-        } else {
-            Ok(())
-        }
-    }
-}
-
-fn do_gettime() -> Result<Timespec, Errno> {
-    let mut time = timespec {
-        tv_sec: 0,
-        tv_nsec: 0,
-    };
-
-    unsafe {
-        if clock_gettime(CLOCK_BOOTTIME, &mut time) == -1 {
-            Err(errno())
-        } else {
-            Ok(time.into())
-        }
-    }
-}
+use unix::{
+    do_chdir, do_execv, do_gettime, do_mkdir, do_mount, do_pivot_root, do_rmdir, do_umount,
+};
 
 fn replace_root() -> Result<(), ()> {
     match do_mount(None, Some("/"), None, MS_REC | MS_PRIVATE, None) {
         Err(e) => {
-            eprintln!("remount \"/\" with private option failed: {}", e);
+            kprintln!("remount \"/\" with private option failed: {}", e);
             return Err(());
         }
-        Ok(_) => eprintln!("\"/\" remounted"),
+        Ok(_) => kprintln!("\"/\" remounted"),
     };
 
     match do_mount(
@@ -173,10 +36,10 @@ fn replace_root() -> Result<(), ()> {
         None,
     ) {
         Err(e) => {
-            eprintln!("mount \"/run\" failed: {}", e);
+            kprintln!("mount \"/run\" failed: {}", e);
             return Err(());
         }
-        Ok(_) => eprintln!("\"/run\" mounted"),
+        Ok(_) => kprintln!("\"/run\" mounted"),
     };
 
     let ovr_dir_structures = [
@@ -186,13 +49,14 @@ fn replace_root() -> Result<(), ()> {
         ("/run/overlay/work", 0o0755),
         ("/run/overlay/merged", 0o0755),
     ];
+
     for (path, mode) in ovr_dir_structures {
         match do_mkdir(path, mode) {
             Err(e) => {
-                eprintln!("mkdir \"{}\" failed: {}", path, e);
+                kprintln!("mkdir \"{}\" failed: {}", path, e);
                 return Err(());
             }
-            Ok(_) => eprintln!("\"{}\" created", path),
+            Ok(_) => kprintln!("\"{}\" created", path),
         }
     }
 
@@ -204,10 +68,10 @@ fn replace_root() -> Result<(), ()> {
         None,
     ) {
         Err(e) => {
-            eprintln!("mount \"/run/overlay/lower\" failed: {}", e);
+            kprintln!("mount \"/run/overlay/lower\" failed: {}", e);
             return Err(());
         }
-        Ok(_) => eprintln!("\"/run/overlay/lower\" mounted"),
+        Ok(_) => kprintln!("\"/run/overlay/lower\" mounted"),
     };
 
     match do_mount(
@@ -216,66 +80,70 @@ fn replace_root() -> Result<(), ()> {
          Some("overlay"), MS_RELATIME|MS_LAZYTIME, 
          Some("lowerdir=/run/overlay/lower,upperdir=/run/overlay/upper,workdir=/run/overlay/work,redirect_dir=on,uuid=on,metacopy=on,volatile"),
         ) {
-        Err(e) => {eprintln!("mount \"/run/overlay/merged\" failed: {}", e);
+        Err(e) => {kprintln!("mount \"/run/overlay/merged\" failed: {}", e);
         return Err(());
     },
-        Ok(_) => eprintln!("\"/run/overlay/merged\" mounted"),
+        Ok(_) => kprintln!("\"/run/overlay/merged\" mounted"),
     };
 
     match do_mkdir("/run/overlay/merged/oldroot", 0o0700) {
         Err(e) => {
-            eprintln!("mkdir \"/run/overlay/merged/oldroot\" failed: {}", e);
+            kprintln!("mkdir \"/run/overlay/merged/oldroot\" failed: {}", e);
             return Err(());
         }
-        Ok(_) => eprintln!("\"/run/overlay/merged/oldroot\" created"),
+        Ok(_) => kprintln!("\"/run/overlay/merged/oldroot\" created"),
     }
 
-    match pivot_root("/run/overlay/merged", "/run/overlay/merged/oldroot") {
+    match do_pivot_root("/run/overlay/merged", "/run/overlay/merged/oldroot") {
         Err(e) => {
-            eprintln!("pivot_root failed: {}", e);
+            kprintln!("pivot_root failed: {}", e);
             return Err(());
         }
-        Ok(_) => eprintln!("pivot_root done"),
+        Ok(_) => kprintln!("pivot_root done"),
     }
 
     match do_chdir("/") {
         Err(e) => {
-            eprintln!("chdir \"/\" failed: {}", e);
+            kprintln!("chdir \"/\" failed: {}", e);
             return Err(());
         }
-        Ok(_) => eprintln!("chdir \"/\" done"),
+        Ok(_) => kprintln!("chdir \"/\" done"),
     }
 
-    match do_mount(
-        Some("/oldroot/run"),
-        Some("/run"),
-        None,
-        MS_REC | MS_MOVE,
-        None,
-    ) {
-        Err(e) => {
-            eprintln!("moving mountpoint \"/oldroot/run\"  failed: {}", e);
-            return Err(());
-        }
-        Ok(_) => eprintln!("mountpoint \"/oldroot/run\" moved"),
-    };
+    let move_mnt_pairs = [
+        ("/oldroot/run", "/run", true),
+        ("/oldroot/dev", "/dev", false),
+    ];
+    for (src, dst, required) in move_mnt_pairs {
+        match (
+            do_mount(Some(src), Some(dst), None, MS_REC | MS_MOVE, None),
+            required,
+        ) {
+            (Err(Errno(EINVAL)), true) => (),
+            (Err(e), _) => {
+                kprintln!("moving mountpoint \"{}\"  failed: {}", src, e);
+                return Err(());
+            }
+            (Ok(_), _) => kprintln!("mountpoint \"{}\" moved", src),
+        };
+    }
 
     match do_umount("/oldroot", MNT_DETACH) {
         Err(e) => {
-            eprintln!("umount mountpoint \"/oldroot\"  failed: {}", e);
+            kprintln!("umount mountpoint \"/oldroot\"  failed: {}", e);
             return Err(());
         }
-        Ok(_) => eprintln!("mountpoint \"/oldroot/run\" unmounted"),
+        Ok(_) => kprintln!("mountpoint \"/oldroot/run\" unmounted"),
     };
 
     let unused_dirs = ["/oldroot", "/run/overlay/merged"];
     for path in unused_dirs {
         match do_rmdir(path) {
             Err(e) => {
-                eprintln!("remove \"{}\" failed: {}", path, e);
+                kprintln!("remove \"{}\" failed: {}", path, e);
                 return Err(());
             }
-            Ok(_) => eprintln!("\"{}\" removed", path),
+            Ok(_) => kprintln!("\"{}\" removed", path),
         }
     }
 
@@ -284,9 +152,10 @@ fn replace_root() -> Result<(), ()> {
 
 #[no_mangle]
 unsafe extern "C" fn main(_argc: c_int, argv: *mut *const u8) -> c_int {
+    init_kmsg();
     let start = match do_gettime() {
         Err(err) => {
-            eprintln!("{}", err);
+            kprintln!("{}", err);
             return EXIT_FAILURE;
         }
         Ok(v) => v,
@@ -294,23 +163,34 @@ unsafe extern "C" fn main(_argc: c_int, argv: *mut *const u8) -> c_int {
 
     match replace_root() {
         Err(_) => return EXIT_FAILURE,
-        Ok(_) => eprintln!("rootfs replaced with overlayfs!"),
+        Ok(_) => kprintln!("rootfs replaced with overlayfs!"),
     };
 
     let end = match do_gettime() {
         Err(err) => {
-            eprintln!("{}", err);
+            kprintln!("{}", err);
             return EXIT_FAILURE;
         }
         Ok(v) => v,
     };
     let elapsed = end - start;
-    eprintln!("processed in {:?}", elapsed);
+    kprintln!("processed in {:?}", elapsed);
 
     // overwrite arg0
-    let init_path = CStr::from_bytes_with_nul_unchecked(b"/sbin/init\0");
-    *argv = init_path.as_ptr();
-    execv(init_path.as_ptr(), argv);
 
-    EXIT_SUCCESS
+    let init_candidates = ["/sbin/init", "/usr/sbin/init", "/usr/lib/systemd/systemd"];
+    for init_path in init_candidates {
+        kprintln!("trying to execute \"{}\"", init_path);
+        match do_execv(init_path, argv) {
+            Err(Errno(ENOENT)) | Err(Errno(EACCES)) | Err(Errno(ENOEXEC)) | Err(Errno(ENOTDIR)) => {
+                continue;
+            }
+            Err(e) => {
+                kprintln!("execution \"{}\" failed: {}", init_path, e);
+                break;
+            }
+            Ok(()) => return EXIT_SUCCESS,
+        };
+    }
+    EXIT_FAILURE
 }

src/mem.rs

@@ -1,18 +1,96 @@
 use core::alloc::{GlobalAlloc, Layout};
-use libc::{c_void, free, malloc};
+use core::{cmp, mem, ptr};
+use libc::{c_void, calloc, free, malloc, posix_memalign, realloc};
+
+const MIN_ALIGN: usize = 16;
+
 /// The global allocator type.
 #[derive(Default)]
 pub struct Allocator;
 
+impl Allocator {}
 unsafe impl GlobalAlloc for Allocator {
+    #[inline]
     unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
-        malloc(layout.size()) as *mut u8
+        // jemalloc provides alignment less than MIN_ALIGN for small allocations.
+        // So only rely on MIN_ALIGN if size >= align.
+        // Also see <https://github.com/rust-lang/rust/issues/45955> and
+        // <https://github.com/rust-lang/rust/issues/62251#issuecomment-507580914>.
+        if layout.align() <= MIN_ALIGN && layout.align() <= layout.size() {
+            malloc(layout.size()) as *mut u8
+        } else {
+            #[cfg(target_os = "macos")]
+            {
+                if layout.align() > (1 << 31) {
+                    return ptr::null_mut();
+                }
+            }
+            aligned_malloc(&layout)
+        }
     }
+
+    #[inline]
+    unsafe fn alloc_zeroed(&self, layout: Layout) -> *mut u8 {
+        // See the comment above in `alloc` for why this check looks the way it does.
+        if layout.align() <= MIN_ALIGN && layout.align() <= layout.size() {
+            calloc(layout.size(), 1) as *mut u8
+        } else {
+            let ptr = self.alloc(layout);
+            if !ptr.is_null() {
+                ptr::write_bytes(ptr, 0, layout.size());
+            }
+            ptr
+        }
+    }
+
+    #[inline]
     unsafe fn dealloc(&self, ptr: *mut u8, _layout: Layout) {
-        free(ptr as *mut c_void);
+        free(ptr as *mut c_void)
+    }
+
+    #[inline]
+    unsafe fn realloc(&self, ptr: *mut u8, layout: Layout, new_size: usize) -> *mut u8 {
+        if layout.align() <= MIN_ALIGN && layout.align() <= new_size {
+            realloc(ptr as *mut c_void, new_size) as *mut u8
+        } else {
+            realloc_fallback(self, ptr, layout, new_size)
+        }
     }
 }
 
 /// The static global allocator.
 #[global_allocator]
 static GLOBAL_ALLOCATOR: Allocator = Allocator;
+
+#[inline]
+unsafe fn aligned_malloc(layout: &Layout) -> *mut u8 {
+    let mut out = ptr::null_mut();
+    // posix_memalign requires that the alignment be a multiple of `sizeof(void*)`.
+    // Since these are all powers of 2, we can just use max.
+    let align = layout.align().max(mem::size_of::<usize>());
+    let ret = posix_memalign(&mut out, align, layout.size());
+    if ret != 0 {
+        ptr::null_mut()
+    } else {
+        out as *mut u8
+    }
+}
+
+#[inline]
+unsafe fn realloc_fallback(
+    allocator: &Allocator,
+    ptr: *mut u8,
+    old_layout: Layout,
+    new_size: usize,
+) -> *mut u8 {
+    // Docs for GlobalAlloc::realloc require this to be valid:
+    let new_layout = Layout::from_size_align_unchecked(new_size, old_layout.align());
+
+    let new_ptr = allocator.alloc(new_layout);
+    if !new_ptr.is_null() {
+        let size = cmp::min(old_layout.size(), new_size);
+        ptr::copy_nonoverlapping(ptr, new_ptr, size);
+        allocator.dealloc(ptr, old_layout);
+    }
+    new_ptr
+}

src/tm.rs

@@ -0,0 +1,42 @@
+use core::ops::Sub;
+use core::time::Duration;
+
+use libc::timespec;
+
+pub(crate) struct Timespec {
+    pub(crate) ts: timespec,
+}
+
+pub(crate) fn new_timespec() -> Timespec {
+    Timespec {
+        ts: timespec {
+            tv_sec: 0,
+            tv_nsec: 0,
+        },
+    }
+}
+impl Sub<Timespec> for Timespec {
+    type Output = Duration;
+
+    fn sub(self, other: Timespec) -> Duration {
+        let sec = self.ts.tv_sec - other.ts.tv_sec;
+        let nsec = self.ts.tv_nsec - other.ts.tv_nsec;
+        if nsec < 0 {
+            Duration::from_secs(sec as u64) - Duration::from_nanos(nsec.unsigned_abs())
+        } else {
+            Duration::from_secs(sec as u64) + Duration::from_nanos(nsec.unsigned_abs())
+        }
+    }
+}
+
+impl From<timespec> for Timespec {
+    fn from(item: timespec) -> Self {
+        Timespec { ts: item }
+    }
+}
+
+impl Into<timespec> for Timespec {
+    fn into(self) -> timespec {
+        self.ts
+    }
+}

src/unix.rs

@@ -0,0 +1,194 @@
+use crate::tm::{new_timespec, Timespec};
+use alloc::ffi::CString;
+use core::borrow::BorrowMut;
+use core::ptr::null;
+use errno::{errno, Errno};
+use libc::{
+    c_void, chdir, clock_gettime, execv, mkdir, mode_t, mount, rmdir, syscall, umount2,
+    SYS_pivot_root, CLOCK_BOOTTIME,
+};
+
+#[cfg(test)]
+use libc::{
+    c_uint, makedev, mkdir, mknod, mode_t, mount, readlinkat, rmdir, strlen, syscall, umount2,
+    unlink, SYS_pivot_root, AT_FDCWD, CLOCK_BOOTTIME, S_IFCHR,
+};
+
+pub(crate) type SystemResult = Result<(), Errno>;
+
+#[cfg(test)]
+pub(crate) fn do_mknod(path: &str, major: c_uint, minor: c_uint) -> SystemResult {
+    // has ownership
+    let raw_path = CString::new(path).unwrap();
+
+    unsafe {
+        let dev = makedev(major, minor);
+
+        if mknod(raw_path.as_ref().as_ptr(), S_IFCHR, dev) == -1 {
+            Err(errno())
+        } else {
+            Ok(())
+        }
+    }
+}
+
+pub(crate) fn do_mount(
+    source: Option<&str>,
+    target: Option<&str>,
+    fs: Option<&str>,
+    flags: u64,
+    opt: Option<&str>,
+) -> SystemResult {
+    // has ownership
+    let raw_src = source.map(|v| CString::new(v).ok()).flatten();
+    let raw_tgt = target.map(|v| CString::new(v).ok()).flatten();
+    let raw_fs = fs.map(|v| CString::new(v).ok()).flatten();
+    let raw_fs_opt = opt.map(|v| CString::new(v).ok()).flatten();
+
+    unsafe {
+        if mount(
+            raw_src.as_ref().map(|v| v.as_ptr()).unwrap_or_else(null),
+            raw_tgt.as_ref().map(|v| v.as_ptr()).unwrap_or_else(null),
+            raw_fs.as_ref().map(|v| v.as_ptr()).unwrap_or_else(null),
+            flags,
+            raw_fs_opt.as_ref().map(|v| v.as_ptr()).unwrap_or_else(null) as *const c_void,
+        ) == -1
+        {
+            Err(errno())
+        } else {
+            Ok(())
+        }
+    }
+}
+
+pub(crate) fn do_umount(path: &str, flags: i32) -> SystemResult {
+    // has ownership
+    let raw_path = CString::new(path).unwrap();
+    unsafe {
+        if umount2(raw_path.as_ref().as_ptr(), flags) == -1 {
+            Err(errno())
+        } else {
+            Ok(())
+        }
+    }
+}
+
+#[cfg(test)]
+pub(crate) fn do_readlink(path: &str) -> Result<CString, Errno> {
+    // has ownership
+    let raw_path = CString::new(path).unwrap();
+    unsafe {
+        const BUF_SZ: usize = 1024;
+        let mut buf = [0u8; BUF_SZ];
+        if readlinkat(
+            AT_FDCWD,
+            raw_path.as_ref().as_ptr(),
+            buf.as_mut_ptr() as *mut _,
+            buf.len() as size_t,
+        ) == -1
+        {
+            return Err(errno());
+        }
+        let c_str_len = strlen(buf.as_ptr() as *const _);
+        Ok(from_utf8_lossy(&buf[..c_str_len]))
+    }
+}
+
+pub(crate) fn do_mkdir(path: &str, mode: mode_t) -> SystemResult {
+    // has ownership
+    let raw_path = CString::new(path).unwrap();
+    unsafe {
+        if mkdir(raw_path.as_ref().as_ptr(), mode) == -1 {
+            Err(errno())
+        } else {
+            Ok(())
+        }
+    }
+}
+
+#[cfg(test)]
+pub(crate) fn do_unlink(path: &str) -> SystemResult {
+    // has ownership
+    let raw_path = CString::new(path).unwrap();
+    unsafe {
+        if unlink(raw_path.as_ref().as_ptr()) == -1 {
+            Err(errno())
+        } else {
+            Ok(())
+        }
+    }
+}
+
+pub(crate) fn do_rmdir(path: &str) -> SystemResult {
+    // has ownership
+    let raw_path = CString::new(path).unwrap();
+    unsafe {
+        if rmdir(raw_path.as_ref().as_ptr()) == -1 {
+            Err(errno())
+        } else {
+            Ok(())
+        }
+    }
+}
+
+pub(crate) fn do_execv(path: &str, argv: *mut *const u8) -> SystemResult {
+    let raw_init_path = CString::new(path).unwrap();
+    unsafe {
+        *argv = raw_init_path.as_ref().as_ptr();
+        if execv(raw_init_path.as_ref().as_ptr(), argv) == -1 {
+            Err(errno())
+        } else {
+            Ok(())
+        }
+    }
+}
+
+pub(crate) fn do_chdir(path: &str) -> SystemResult {
+    // has ownership
+    let raw_path = CString::new(path).unwrap();
+    unsafe {
+        if chdir(raw_path.as_ref().as_ptr()) == -1 {
+            Err(errno())
+        } else {
+            Ok(())
+        }
+    }
+}
+
+pub(crate) fn do_pivot_root(new_root: &str, put_old: &str) -> SystemResult {
+    let raw_new_root = CString::new(new_root).unwrap();
+    let raw_put_old = CString::new(put_old).unwrap();
+    unsafe {
+        if syscall(
+            SYS_pivot_root,
+            raw_new_root.as_ref().as_ptr(),
+            raw_put_old.as_ref().as_ptr(),
+        ) == -1
+        {
+            Err(errno())
+        } else {
+            Ok(())
+        }
+    }
+}
+
+pub(crate) fn do_gettime() -> Result<Timespec, Errno> {
+    let mut time = new_timespec();
+
+    unsafe {
+        if clock_gettime(CLOCK_BOOTTIME, time.ts.borrow_mut()) == -1 {
+            Err(errno())
+        } else {
+            Ok(time.into())
+        }
+    }
+}
+
+#[cfg(test)]
+#[inline]
+fn from_utf8_lossy(input: &[u8]) -> &str {
+    match str::from_utf8(input) {
+        Ok(valid) => valid,
+        Err(error) => unsafe { str::from_utf8_unchecked(&input[..error.valid_up_to()]) },
+    }
+}