blob: ea1017957c7a7fef82473e1c804d7bbe07271d20 [file] [log] [blame]
use crate::fd::{AsFd, BorrowedFd, OwnedFd};
use crate::ffi::{CStr, CString};
use crate::fs::{
fcntl_getfl, fstat, fstatfs, fstatvfs, openat, FileType, Mode, OFlags, Stat, StatFs, StatVfs,
};
use crate::io;
#[cfg(feature = "process")]
use crate::process::fchdir;
use crate::utils::as_ptr;
use alloc::borrow::ToOwned;
use alloc::vec::Vec;
use core::fmt;
use core::mem::size_of;
use linux_raw_sys::general::{linux_dirent64, SEEK_SET};
/// `DIR*`
pub struct Dir {
/// The `OwnedFd` that we read directory entries from.
fd: OwnedFd,
/// Have we seen any errors in this iteration?
any_errors: bool,
/// Should we rewind the stream on the next iteration?
rewind: bool,
/// The buffer for `linux_dirent64` entries.
buf: Vec<u8>,
/// Where we are in the buffer.
pos: usize,
}
impl Dir {
/// Construct a `Dir` that reads entries from the given directory
/// file descriptor.
#[inline]
pub fn read_from<Fd: AsFd>(fd: Fd) -> io::Result<Self> {
Self::_read_from(fd.as_fd())
}
#[inline]
fn _read_from(fd: BorrowedFd<'_>) -> io::Result<Self> {
let flags = fcntl_getfl(fd)?;
let fd_for_dir = openat(fd, cstr!("."), flags | OFlags::CLOEXEC, Mode::empty())?;
Ok(Self {
fd: fd_for_dir,
any_errors: false,
rewind: false,
buf: Vec::new(),
pos: 0,
})
}
/// `rewinddir(self)`
#[inline]
pub fn rewind(&mut self) {
self.any_errors = false;
self.rewind = true;
self.pos = self.buf.len();
}
/// `readdir(self)`, where `None` means the end of the directory.
pub fn read(&mut self) -> Option<io::Result<DirEntry>> {
// If we've seen errors, don't continue to try to read anyting further.
if self.any_errors {
return None;
}
// If a rewind was requested, seek to the beginning.
if self.rewind {
self.rewind = false;
match io::retry_on_intr(|| {
crate::backend::fs::syscalls::_seek(self.fd.as_fd(), 0, SEEK_SET)
}) {
Ok(_) => (),
Err(err) => {
self.any_errors = true;
return Some(Err(err));
}
}
}
// Compute linux_dirent64 field offsets.
let z = linux_dirent64 {
d_ino: 0_u64,
d_off: 0_i64,
d_type: 0_u8,
d_reclen: 0_u16,
d_name: Default::default(),
};
let base = as_ptr(&z) as usize;
let offsetof_d_reclen = (as_ptr(&z.d_reclen) as usize) - base;
let offsetof_d_name = (as_ptr(&z.d_name) as usize) - base;
let offsetof_d_ino = (as_ptr(&z.d_ino) as usize) - base;
let offsetof_d_type = (as_ptr(&z.d_type) as usize) - base;
// Test if we need more entries, and if so, read more.
if self.buf.len() - self.pos < size_of::<linux_dirent64>() {
match self.read_more()? {
Ok(()) => (),
Err(err) => return Some(Err(err)),
}
}
// We successfully read an entry. Extract the fields.
let pos = self.pos;
// Do an unaligned u16 load.
let d_reclen = u16::from_ne_bytes([
self.buf[pos + offsetof_d_reclen],
self.buf[pos + offsetof_d_reclen + 1],
]);
assert!(self.buf.len() - pos >= d_reclen as usize);
self.pos += d_reclen as usize;
// Read the NUL-terminated name from the `d_name` field. Without
// `unsafe`, we need to scan for the NUL twice: once to obtain a size
// for the slice, and then once within `CStr::from_bytes_with_nul`.
let name_start = pos + offsetof_d_name;
let name_len = self.buf[name_start..]
.iter()
.position(|x| *x == b'\0')
.unwrap();
let name = CStr::from_bytes_with_nul(&self.buf[name_start..][..=name_len]).unwrap();
let name = name.to_owned();
assert!(name.as_bytes().len() <= self.buf.len() - name_start);
// Do an unaligned u64 load.
let d_ino = u64::from_ne_bytes([
self.buf[pos + offsetof_d_ino],
self.buf[pos + offsetof_d_ino + 1],
self.buf[pos + offsetof_d_ino + 2],
self.buf[pos + offsetof_d_ino + 3],
self.buf[pos + offsetof_d_ino + 4],
self.buf[pos + offsetof_d_ino + 5],
self.buf[pos + offsetof_d_ino + 6],
self.buf[pos + offsetof_d_ino + 7],
]);
let d_type = self.buf[pos + offsetof_d_type];
// Check that our types correspond to the `linux_dirent64` types.
let _ = linux_dirent64 {
d_ino,
d_off: 0,
d_type,
d_reclen,
d_name: Default::default(),
};
Some(Ok(DirEntry {
d_ino,
d_type,
name,
}))
}
fn read_more(&mut self) -> Option<io::Result<()>> {
// The first few times we're called, we allocate a relatively small
// buffer, because many directories are small. If we're called more,
// use progressively larger allocations, up to a fixed maximum.
//
// The specific sizes and policy here have not been tuned in detail yet
// and may need to be adjusted. In doing so, we should be careful to
// avoid unbounded buffer growth. This buffer only exists to share the
// cost of a `getdents` call over many entries, so if it gets too big,
// cache and heap usage will outweigh the benefit. And ultimately,
// directories can contain more entries than we can allocate contiguous
// memory for, so we'll always need to cap the size at some point.
if self.buf.len() < 1024 * size_of::<linux_dirent64>() {
self.buf.reserve(32 * size_of::<linux_dirent64>());
}
self.buf.resize(self.buf.capacity(), 0);
let nread = match io::retry_on_intr(|| {
crate::backend::fs::syscalls::getdents(self.fd.as_fd(), &mut self.buf)
}) {
Ok(nread) => nread,
Err(io::Errno::NOENT) => {
self.any_errors = true;
return None;
}
Err(err) => {
self.any_errors = true;
return Some(Err(err));
}
};
self.buf.resize(nread, 0);
self.pos = 0;
if nread == 0 {
None
} else {
Some(Ok(()))
}
}
/// `fstat(self)`
#[inline]
pub fn stat(&self) -> io::Result<Stat> {
fstat(&self.fd)
}
/// `fstatfs(self)`
#[inline]
pub fn statfs(&self) -> io::Result<StatFs> {
fstatfs(&self.fd)
}
/// `fstatvfs(self)`
#[inline]
pub fn statvfs(&self) -> io::Result<StatVfs> {
fstatvfs(&self.fd)
}
/// `fchdir(self)`
#[cfg(feature = "process")]
#[cfg_attr(doc_cfg, doc(cfg(feature = "process")))]
#[inline]
pub fn chdir(&self) -> io::Result<()> {
fchdir(&self.fd)
}
}
impl Iterator for Dir {
type Item = io::Result<DirEntry>;
#[inline]
fn next(&mut self) -> Option<Self::Item> {
Self::read(self)
}
}
impl fmt::Debug for Dir {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.debug_struct("Dir").field("fd", &self.fd).finish()
}
}
/// `struct dirent`
#[derive(Debug)]
pub struct DirEntry {
d_ino: u64,
d_type: u8,
name: CString,
}
impl DirEntry {
/// Returns the file name of this directory entry.
#[inline]
pub fn file_name(&self) -> &CStr {
&self.name
}
/// Returns the type of this directory entry.
#[inline]
pub fn file_type(&self) -> FileType {
FileType::from_dirent_d_type(self.d_type)
}
/// Return the inode number of this directory entry.
#[inline]
pub fn ino(&self) -> u64 {
self.d_ino
}
}
#[test]
fn dir_iterator_handles_io_errors() {
// create a dir, keep the FD, then delete the dir
let tmp = tempfile::tempdir().unwrap();
let fd = crate::fs::openat(
crate::fs::CWD,
tmp.path(),
crate::fs::OFlags::RDONLY | crate::fs::OFlags::CLOEXEC,
crate::fs::Mode::empty(),
)
.unwrap();
let file_fd = crate::fs::openat(
&fd,
tmp.path().join("test.txt"),
crate::fs::OFlags::WRONLY | crate::fs::OFlags::CREATE,
crate::fs::Mode::RWXU,
)
.unwrap();
let mut dir = Dir::read_from(&fd).unwrap();
// Reach inside the `Dir` and replace its directory with a file, which
// will cause the subsequent `getdents64` to fail.
crate::io::dup2(&file_fd, &mut dir.fd).unwrap();
assert!(matches!(dir.next(), Some(Err(_))));
assert!(matches!(dir.next(), None));
}