use crate::{ Error, Errno, NixPath, Result, sys::uio::IoVec }; use libc::{c_char, c_int, c_uint, c_void}; use std::{ borrow::Cow, ffi::{CString, CStr}, fmt, io, ptr }; libc_bitflags!( /// Used with [`Nmount::nmount`]. pub struct MntFlags: c_int { /// ACL support enabled. #[cfg(any(target_os = "netbsd", target_os = "freebsd"))] MNT_ACLS; /// All I/O to the file system should be done asynchronously. MNT_ASYNC; /// dir should instead be a file system ID encoded as “FSID:val0:val1”. #[cfg(target_os = "freebsd")] MNT_BYFSID; /// Force a read-write mount even if the file system appears to be /// unclean. MNT_FORCE; /// GEOM journal support enabled. #[cfg(target_os = "freebsd")] MNT_GJOURNAL; /// MAC support for objects. #[cfg(any(target_os = "macos", target_os = "freebsd"))] MNT_MULTILABEL; /// Disable read clustering. #[cfg(any(target_os = "dragonfly", target_os = "freebsd"))] MNT_NOCLUSTERR; /// Disable write clustering. #[cfg(any(target_os = "dragonfly", target_os = "freebsd"))] MNT_NOCLUSTERW; /// Enable NFS version 4 ACLs. #[cfg(target_os = "freebsd")] MNT_NFS4ACLS; /// Do not update access times. MNT_NOATIME; /// Disallow program execution. MNT_NOEXEC; /// Do not honor setuid or setgid bits on files when executing them. MNT_NOSUID; /// Do not follow symlinks. #[cfg(any(target_os = "dragonfly", target_os = "freebsd"))] MNT_NOSYMFOLLOW; /// Mount read-only. MNT_RDONLY; /// Causes the vfs subsystem to update its data structures pertaining to /// the specified already mounted file system. MNT_RELOAD; /// Create a snapshot of the file system. /// /// See [mksnap_ffs(8)](https://www.freebsd.org/cgi/man.cgi?query=mksnap_ffs) #[cfg(any(target_os = "macos", target_os = "freebsd"))] MNT_SNAPSHOT; /// Using soft updates. #[cfg(any( target_os = "dragonfly", target_os = "freebsd", target_os = "netbsd", target_os = "openbsd" ))] MNT_SOFTDEP; /// Directories with the SUID bit set chown new files to their own /// owner. #[cfg(any(target_os = "dragonfly", target_os = "freebsd"))] MNT_SUIDDIR; /// All I/O to the file system should be done synchronously. MNT_SYNCHRONOUS; /// Union with underlying fs. #[cfg(any( target_os = "macos", target_os = "freebsd", target_os = "netbsd" ))] MNT_UNION; /// Indicates that the mount command is being applied to an already /// mounted file system. MNT_UPDATE; /// Check vnode use counts. #[cfg(target_os = "freebsd")] MNT_NONBUSY; } ); /// The Error type of [`Nmount::nmount`]. /// /// It wraps an [`Errno`], but also may contain an additional message returned /// by `nmount(2)`. #[derive(Debug)] pub struct NmountError { errno: Error, errmsg: Option } impl NmountError { /// Returns the additional error string sometimes generated by `nmount(2)`. pub fn errmsg(&self) -> Option<&str> { self.errmsg.as_deref() } /// Returns the inner [`Error`] pub const fn error(&self) -> Error { self.errno } fn new(error: Error, errmsg: Option<&CStr>) -> Self { Self { errno: error, errmsg: errmsg.map(CStr::to_string_lossy).map(Cow::into_owned) } } } impl std::error::Error for NmountError {} impl fmt::Display for NmountError { fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { if let Some(errmsg) = &self.errmsg { write!(f, "{:?}: {}: {}", self.errno, errmsg, self.errno.desc()) } else { write!(f, "{:?}: {}", self.errno, self.errno.desc()) } } } impl From for io::Error { fn from(err: NmountError) -> Self { err.errno.into() } } /// Result type of [`Nmount::nmount`]. pub type NmountResult = std::result::Result<(), NmountError>; /// Mount a FreeBSD file system. /// /// The `nmount(2)` system call works similarly to the `mount(8)` program; it /// takes its options as a series of name-value pairs. Most of the values are /// strings, as are all of the names. The `Nmount` structure builds up an /// argument list and then executes the syscall. /// /// # Examples /// /// To mount `target` onto `mountpoint` with `nullfs`: /// ``` /// # use nix::unistd::Uid; /// # use ::sysctl::CtlValue; /// # if !Uid::current().is_root() && CtlValue::Int(0) == ::sysctl::value("vfs.usermount").unwrap() { /// # return; /// # }; /// use nix::mount::{MntFlags, Nmount, unmount}; /// use std::ffi::CString; /// use tempfile::tempdir; /// /// let mountpoint = tempdir().unwrap(); /// let target = tempdir().unwrap(); /// /// let fstype = CString::new("fstype").unwrap(); /// let nullfs = CString::new("nullfs").unwrap(); /// Nmount::new() /// .str_opt(&fstype, &nullfs) /// .str_opt_owned("fspath", mountpoint.path().to_str().unwrap()) /// .str_opt_owned("target", target.path().to_str().unwrap()) /// .nmount(MntFlags::empty()).unwrap(); /// /// unmount(mountpoint.path(), MntFlags::empty()).unwrap(); /// ``` /// /// # See Also /// * [`nmount(2)`](https://www.freebsd.org/cgi/man.cgi?query=nmount) /// * [`nullfs(5)`](https://www.freebsd.org/cgi/man.cgi?query=nullfs) #[cfg(target_os = "freebsd")] #[derive(Debug, Default)] pub struct Nmount<'a>{ iov: Vec>, is_owned: Vec, } #[cfg(target_os = "freebsd")] impl<'a> Nmount<'a> { /// Add an opaque mount option. /// /// Some file systems take binary-valued mount options. They can be set /// with this method. /// /// # Safety /// /// Unsafe because it will cause `Nmount::nmount` to dereference a raw /// pointer. The user is responsible for ensuring that `val` is valid and /// its lifetime outlives `self`! An easy way to do that is to give the /// value a larger scope than `name` /// /// # Examples /// ``` /// use libc::c_void; /// use nix::mount::Nmount; /// use std::ffi::CString; /// use std::mem; /// /// // Note that flags outlives name /// let mut flags: u32 = 0xdeadbeef; /// let name = CString::new("flags").unwrap(); /// let p = &mut flags as *mut u32 as *mut c_void; /// let len = mem::size_of_val(&flags); /// let mut nmount = Nmount::new(); /// unsafe { nmount.mut_ptr_opt(&name, p, len) }; /// ``` pub unsafe fn mut_ptr_opt( &mut self, name: &'a CStr, val: *mut c_void, len: usize ) -> &mut Self { self.iov.push(IoVec::from_slice(name.to_bytes_with_nul())); self.is_owned.push(false); self.iov.push(IoVec::from_raw_parts(val, len)); self.is_owned.push(false); self } /// Add a mount option that does not take a value. /// /// # Examples /// ``` /// use nix::mount::Nmount; /// use std::ffi::CString; /// /// let read_only = CString::new("ro").unwrap(); /// Nmount::new() /// .null_opt(&read_only); /// ``` pub fn null_opt(&mut self, name: &'a CStr) -> &mut Self { self.iov.push(IoVec::from_slice(name.to_bytes_with_nul())); self.is_owned.push(false); self.iov.push(IoVec::from_raw_parts(ptr::null_mut(), 0)); self.is_owned.push(false); self } /// Add a mount option that does not take a value, but whose name must be /// owned. /// /// /// This has higher runtime cost than [`Nmount::null_opt`], but is useful /// when the name's lifetime doesn't outlive the `Nmount`, or it's a /// different string type than `CStr`. /// /// # Examples /// ``` /// use nix::mount::Nmount; /// /// let read_only = "ro"; /// let mut nmount: Nmount<'static> = Nmount::new(); /// nmount.null_opt_owned(read_only); /// ``` pub fn null_opt_owned(&mut self, name: &P) -> &mut Self { name.with_nix_path(|s| { let len = s.to_bytes_with_nul().len(); self.iov.push(IoVec::from_raw_parts( // Must free it later s.to_owned().into_raw() as *mut c_void, len )); self.is_owned.push(true); }).unwrap(); self.iov.push(IoVec::from_raw_parts(ptr::null_mut(), 0)); self.is_owned.push(false); self } /// Add a mount option as a [`CStr`]. /// /// # Examples /// ``` /// use nix::mount::Nmount; /// use std::ffi::CString; /// /// let fstype = CString::new("fstype").unwrap(); /// let nullfs = CString::new("nullfs").unwrap(); /// Nmount::new() /// .str_opt(&fstype, &nullfs); /// ``` pub fn str_opt( &mut self, name: &'a CStr, val: &'a CStr ) -> &mut Self { self.iov.push(IoVec::from_slice(name.to_bytes_with_nul())); self.is_owned.push(false); self.iov.push(IoVec::from_slice(val.to_bytes_with_nul())); self.is_owned.push(false); self } /// Add a mount option as an owned string. /// /// This has higher runtime cost than [`Nmount::str_opt`], but is useful /// when the value's lifetime doesn't outlive the `Nmount`, or it's a /// different string type than `CStr`. /// /// # Examples /// ``` /// use nix::mount::Nmount; /// use std::path::Path; /// /// let mountpoint = Path::new("/mnt"); /// Nmount::new() /// .str_opt_owned("fspath", mountpoint.to_str().unwrap()); /// ``` pub fn str_opt_owned(&mut self, name: &P1, val: &P2) -> &mut Self where P1: ?Sized + NixPath, P2: ?Sized + NixPath { name.with_nix_path(|s| { let len = s.to_bytes_with_nul().len(); self.iov.push(IoVec::from_raw_parts( // Must free it later s.to_owned().into_raw() as *mut c_void, len )); self.is_owned.push(true); }).unwrap(); val.with_nix_path(|s| { let len = s.to_bytes_with_nul().len(); self.iov.push(IoVec::from_raw_parts( // Must free it later s.to_owned().into_raw() as *mut c_void, len )); self.is_owned.push(true); }).unwrap(); self } pub fn new() -> Self { Self::default() } /// Actually mount the file system. pub fn nmount(&mut self, flags: MntFlags) -> NmountResult { // nmount can return extra error information via a "errmsg" return // argument. const ERRMSG_NAME: &[u8] = b"errmsg\0"; let mut errmsg = vec![0u8; 255]; self.iov.push(IoVec::from_raw_parts( ERRMSG_NAME.as_ptr() as *mut c_void, ERRMSG_NAME.len() )); self.iov.push(IoVec::from_raw_parts( errmsg.as_mut_ptr() as *mut c_void, errmsg.len() )); let niov = self.iov.len() as c_uint; let iovp = self.iov.as_mut_ptr() as *mut libc::iovec; let res = unsafe { libc::nmount(iovp, niov, flags.bits) }; match Errno::result(res) { Ok(_) => Ok(()), Err(error) => { let errmsg = match errmsg.iter().position(|&x| x == 0) { None => None, Some(0) => None, Some(n) => { let sl = &errmsg[0..n + 1]; Some(CStr::from_bytes_with_nul(sl).unwrap()) } }; Err(NmountError::new(error.into(), errmsg)) } } } } #[cfg(target_os = "freebsd")] impl<'a> Drop for Nmount<'a> { fn drop(&mut self) { for (iov, is_owned) in self.iov.iter().zip(self.is_owned.iter()) { if *is_owned { // Free the owned string. Safe because we recorded ownership, // and Nmount does not implement Clone. unsafe { CString::from_raw(iov.0.iov_base as *mut c_char); } } } } } /// Unmount the file system mounted at `mountpoint`. /// /// Useful flags include /// * `MNT_FORCE` - Unmount even if still in use. /// * `MNT_BYFSID` - `mountpoint` is not a path, but a file system ID /// encoded as `FSID:val0:val1`, where `val0` and `val1` /// are the contents of the `fsid_t val[]` array in decimal. /// The file system that has the specified file system ID /// will be unmounted. See /// [`statfs`](crate::sys::statfs::statfs) to determine the /// `fsid`. pub fn unmount

(mountpoint: &P, flags: MntFlags) -> Result<()> where P: ?Sized + NixPath { let res = mountpoint.with_nix_path(|cstr| { unsafe { libc::unmount(cstr.as_ptr(), flags.bits) } })?; Errno::result(res).map(drop) }