blob: de26e6f007da77bd67544f163eebed4b448ef800 [file] [log] [blame]
use std::{
ffi::CString,
io,
os::unix::ffi::OsStrExt,
os::unix::io::{
AsRawFd,
FromRawFd,
IntoRawFd,
RawFd,
},
path::Path,
sync::{
atomic::AtomicBool,
Arc,
}
};
use inotify_sys as ffi;
use libc::{
F_GETFL,
F_SETFD,
F_SETFL,
FD_CLOEXEC,
O_NONBLOCK,
fcntl,
};
use crate::events::Events;
use crate::fd_guard::FdGuard;
use crate::util::read_into_buffer;
use crate::watches::{
WatchDescriptor,
WatchMask,
};
#[cfg(feature = "stream")]
use crate::stream::EventStream;
/// Idiomatic Rust wrapper around Linux's inotify API
///
/// `Inotify` is a wrapper around an inotify instance. It generally tries to
/// adhere to the underlying inotify API closely, while making access to it
/// safe and convenient.
///
/// Please refer to the [top-level documentation] for further details and a
/// usage example.
///
/// [top-level documentation]: index.html
pub struct Inotify {
fd: Arc<FdGuard>,
}
impl Inotify {
/// Creates an [`Inotify`] instance
///
/// Initializes an inotify instance by calling [`inotify_init1`].
///
/// This method passes both flags accepted by [`inotify_init1`], not giving
/// the user any choice in the matter, as not passing the flags would be
/// inappropriate in the context of this wrapper:
///
/// - [`IN_CLOEXEC`] prevents leaking file descriptors to other processes.
/// - [`IN_NONBLOCK`] controls the blocking behavior of the inotify API,
/// which is entirely managed by this wrapper.
///
/// # Errors
///
/// Directly returns the error from the call to [`inotify_init1`], without
/// adding any error conditions of its own.
///
/// # Examples
///
/// ```
/// use inotify::Inotify;
///
/// let inotify = Inotify::init()
/// .expect("Failed to initialize an inotify instance");
/// ```
///
/// [`Inotify`]: struct.Inotify.html
/// [`inotify_init1`]: ../inotify_sys/fn.inotify_init1.html
/// [`IN_CLOEXEC`]: ../inotify_sys/constant.IN_CLOEXEC.html
/// [`IN_NONBLOCK`]: ../inotify_sys/constant.IN_NONBLOCK.html
pub fn init() -> io::Result<Inotify> {
// Initialize inotify and set CLOEXEC and NONBLOCK flags.
//
// NONBLOCK is needed, because `Inotify` manages blocking behavior for
// the API consumer, and the way we do that is to make everything non-
// blocking by default and later override that as required.
//
// CLOEXEC prevents leaking file descriptors to processes executed by
// this process and seems to be a best practice. I don't grasp this
// issue completely and failed to find any authoritative sources on the
// topic. There's some discussion in the open(2) and fcntl(2) man pages,
// but I didn't find that helpful in understanding the issue of leaked
// file descriptors. For what it's worth, there's a Rust issue about
// this:
// https://github.com/rust-lang/rust/issues/12148
let fd = unsafe {
let fd = ffi::inotify_init();
if fd == -1 {
return Err(io::Error::last_os_error());
}
if fcntl(fd, F_SETFD, FD_CLOEXEC) == -1 {
return Err(io::Error::last_os_error());
}
if fcntl(fd, F_SETFL, O_NONBLOCK) == -1 {
return Err(io::Error::last_os_error());
}
fd
};
Ok(Inotify {
fd: Arc::new(FdGuard {
fd,
close_on_drop: AtomicBool::new(true),
}),
})
}
/// Adds or updates a watch for the given path
///
/// Adds a new watch or updates an existing one for the file referred to by
/// `path`. Returns a watch descriptor that can be used to refer to this
/// watch later.
///
/// The `mask` argument defines what kind of changes the file should be
/// watched for, and how to do that. See the documentation of [`WatchMask`]
/// for details.
///
/// If this method is used to add a new watch, a new [`WatchDescriptor`] is
/// returned. If it is used to update an existing watch, a
/// [`WatchDescriptor`] that equals the previously returned
/// [`WatchDescriptor`] for that watch is returned instead.
///
/// Under the hood, this method just calls [`inotify_add_watch`] and does
/// some trivial translation between the types on the Rust side and the C
/// side.
///
/// # Attention: Updating watches and hardlinks
///
/// As mentioned above, this method can be used to update an existing watch.
/// This is usually done by calling this method with the same `path`
/// argument that it has been called with before. But less obviously, it can
/// also happen if the method is called with a different path that happens
/// to link to the same inode.
///
/// You can detect this by keeping track of [`WatchDescriptor`]s and the
/// paths they have been returned for. If the same [`WatchDescriptor`] is
/// returned for a different path (and you haven't freed the
/// [`WatchDescriptor`] by removing the watch), you know you have two paths
/// pointing to the same inode, being watched by the same watch.
///
/// # Errors
///
/// Directly returns the error from the call to
/// [`inotify_add_watch`][`inotify_add_watch`] (translated into an
/// `io::Error`), without adding any error conditions of
/// its own.
///
/// # Examples
///
/// ```
/// use inotify::{
/// Inotify,
/// WatchMask,
/// };
///
/// let mut inotify = Inotify::init()
/// .expect("Failed to initialize an inotify instance");
///
/// # // Create a temporary file, so `add_watch` won't return an error.
/// # use std::fs::File;
/// # File::create("/tmp/inotify-rs-test-file")
/// # .expect("Failed to create test file");
/// #
/// inotify.add_watch("/tmp/inotify-rs-test-file", WatchMask::MODIFY)
/// .expect("Failed to add file watch");
///
/// // Handle events for the file here
/// ```
///
/// [`inotify_add_watch`]: ../inotify_sys/fn.inotify_add_watch.html
/// [`WatchMask`]: struct.WatchMask.html
/// [`WatchDescriptor`]: struct.WatchDescriptor.html
pub fn add_watch<P>(&mut self, path: P, mask: WatchMask)
-> io::Result<WatchDescriptor>
where P: AsRef<Path>
{
let path = CString::new(path.as_ref().as_os_str().as_bytes())?;
let wd = unsafe {
ffi::inotify_add_watch(
**self.fd,
path.as_ptr() as *const _,
mask.bits(),
)
};
match wd {
-1 => Err(io::Error::last_os_error()),
_ => Ok(WatchDescriptor{ id: wd, fd: Arc::downgrade(&self.fd) }),
}
}
/// Stops watching a file
///
/// Removes the watch represented by the provided [`WatchDescriptor`] by
/// calling [`inotify_rm_watch`]. [`WatchDescriptor`]s can be obtained via
/// [`Inotify::add_watch`], or from the `wd` field of [`Event`].
///
/// # Errors
///
/// Directly returns the error from the call to [`inotify_rm_watch`].
/// Returns an [`io::Error`] with [`ErrorKind`]`::InvalidInput`, if the given
/// [`WatchDescriptor`] did not originate from this [`Inotify`] instance.
///
/// # Examples
///
/// ```
/// use inotify::Inotify;
///
/// let mut inotify = Inotify::init()
/// .expect("Failed to initialize an inotify instance");
///
/// # // Create a temporary file, so `add_watch` won't return an error.
/// # use std::fs::File;
/// # let mut test_file = File::create("/tmp/inotify-rs-test-file")
/// # .expect("Failed to create test file");
/// #
/// # // Add a watch and modify the file, so the code below doesn't block
/// # // forever.
/// # use inotify::WatchMask;
/// # inotify.add_watch("/tmp/inotify-rs-test-file", WatchMask::MODIFY)
/// # .expect("Failed to add file watch");
/// # use std::io::Write;
/// # write!(&mut test_file, "something\n")
/// # .expect("Failed to write something to test file");
/// #
/// let mut buffer = [0; 1024];
/// let events = inotify
/// .read_events_blocking(&mut buffer)
/// .expect("Error while waiting for events");
///
/// for event in events {
/// inotify.rm_watch(event.wd);
/// }
/// ```
///
/// [`WatchDescriptor`]: struct.WatchDescriptor.html
/// [`inotify_rm_watch`]: ../inotify_sys/fn.inotify_rm_watch.html
/// [`Inotify::add_watch`]: struct.Inotify.html#method.add_watch
/// [`Event`]: struct.Event.html
/// [`Inotify`]: struct.Inotify.html
/// [`io::Error`]: https://doc.rust-lang.org/std/io/struct.Error.html
/// [`ErrorKind`]: https://doc.rust-lang.org/std/io/enum.ErrorKind.html
pub fn rm_watch(&mut self, wd: WatchDescriptor) -> io::Result<()> {
if wd.fd.upgrade().as_ref() != Some(&self.fd) {
return Err(io::Error::new(
io::ErrorKind::InvalidInput,
"Invalid WatchDescriptor",
));
}
let result = unsafe { ffi::inotify_rm_watch(**self.fd, wd.id) };
match result {
0 => Ok(()),
-1 => Err(io::Error::last_os_error()),
_ => panic!(
"unexpected return code from inotify_rm_watch ({})", result)
}
}
/// Waits until events are available, then returns them
///
/// Blocks the current thread until at least one event is available. If this
/// is not desirable, please consider [`Inotify::read_events`].
///
/// This method calls [`Inotify::read_events`] internally and behaves
/// essentially the same, apart from the blocking behavior. Please refer to
/// the documentation of [`Inotify::read_events`] for more information.
///
/// [`Inotify::read_events`]: struct.Inotify.html#method.read_events
/// [`read`]: ../libc/fn.read.html
pub fn read_events_blocking<'a>(&mut self, buffer: &'a mut [u8])
-> io::Result<Events<'a>>
{
unsafe {
let res = fcntl(**self.fd, F_GETFL);
if res == -1 {
return Err(io::Error::last_os_error());
}
if fcntl(**self.fd, F_SETFL, res & !O_NONBLOCK) == -1 {
return Err(io::Error::last_os_error());
}
};
let result = self.read_events(buffer);
unsafe {
let res = fcntl(**self.fd, F_GETFL);
if res == -1 {
return Err(io::Error::last_os_error());
}
if fcntl(**self.fd, F_SETFL, res | O_NONBLOCK) == -1 {
return Err(io::Error::last_os_error());
}
};
result
}
/// Returns one buffer's worth of available events
///
/// Reads as many events as possible into `buffer`, and returns an iterator
/// over them. If no events are available, an iterator is still returned. If
/// you need a method that will block until at least one event is available,
/// please consider [`read_events_blocking`].
///
/// Please note that inotify will merge identical successive unread events
/// into a single event. This means this method can not be used to count the
/// number of file system events.
///
/// The `buffer` argument, as the name indicates, is used as a buffer for
/// the inotify events. Its contents may be overwritten.
///
/// # Errors
///
/// This function directly returns all errors from the call to [`read`]
/// (except EGAIN/EWOULDBLOCK, which result in an empty iterator). In
/// addition, [`ErrorKind::UnexpectedEof`] is returned, if the call to
/// [`read`] returns `0`, signaling end-of-file.
///
/// If `buffer` is too small, this will result in an error with
/// [`ErrorKind::InvalidInput`]. On very old Linux kernels,
/// [`ErrorKind::UnexpectedEof`] will be returned instead.
///
/// # Examples
///
/// ```
/// use inotify::Inotify;
///
/// let mut inotify = Inotify::init()
/// .expect("Failed to initialize an inotify instance");
///
/// let mut buffer = [0; 1024];
/// let events = inotify.read_events(&mut buffer)
/// .expect("Error while reading events");
///
/// for event in events {
/// // Handle event
/// }
/// ```
///
/// [`read_events_blocking`]: struct.Inotify.html#method.read_events_blocking
/// [`read`]: ../libc/fn.read.html
/// [`ErrorKind::UnexpectedEof`]: https://doc.rust-lang.org/std/io/enum.ErrorKind.html#variant.UnexpectedEof
/// [`ErrorKind::InvalidInput`]: https://doc.rust-lang.org/std/io/enum.ErrorKind.html#variant.InvalidInput
pub fn read_events<'a>(&mut self, buffer: &'a mut [u8])
-> io::Result<Events<'a>>
{
let num_bytes = read_into_buffer(**self.fd, buffer);
let num_bytes = match num_bytes {
0 => {
return Err(
io::Error::new(
io::ErrorKind::UnexpectedEof,
"`read` return `0`, signaling end-of-file"
)
);
}
-1 => {
let error = io::Error::last_os_error();
if error.kind() == io::ErrorKind::WouldBlock {
return Ok(Events::new(Arc::downgrade(&self.fd), buffer, 0));
}
else {
return Err(error);
}
},
_ if num_bytes < 0 => {
panic!("{} {} {} {} {} {}",
"Unexpected return value from `read`. Received a negative",
"value that was not `-1`. According to the `read` man page",
"this shouldn't happen, as either `-1` is returned on",
"error, `0` on end-of-file, or a positive value for the",
"number of bytes read. Returned value:",
num_bytes,
);
}
_ => {
// The value returned by `read` should be `isize`. Let's quickly
// verify this with the following assignment, so we can be sure
// our cast below is valid.
let num_bytes: isize = num_bytes;
// The type returned by `read` is `isize`, and we've ruled out
// all negative values with the match arms above. This means we
// can safely cast to `usize`.
debug_assert!(num_bytes > 0);
num_bytes as usize
}
};
Ok(Events::new(Arc::downgrade(&self.fd), buffer, num_bytes))
}
/// Create a stream which collects events
///
/// Returns a `Stream` over all events that are available. This stream is an
/// infinite source of events.
///
/// An internal buffer which can hold the largest possible event is used.
#[cfg(feature = "stream")]
pub fn event_stream<T>(&mut self, buffer: T)
-> io::Result<EventStream<T>>
where
T: AsMut<[u8]> + AsRef<[u8]>,
{
EventStream::new(self.fd.clone(), buffer)
}
/// Closes the inotify instance
///
/// Closes the file descriptor referring to the inotify instance. The user
/// usually doesn't have to call this function, as the underlying inotify
/// instance is closed automatically, when [`Inotify`] is dropped.
///
/// # Errors
///
/// Directly returns the error from the call to [`close`], without adding any
/// error conditions of its own.
///
/// # Examples
///
/// ```
/// use inotify::Inotify;
///
/// let mut inotify = Inotify::init()
/// .expect("Failed to initialize an inotify instance");
///
/// inotify.close()
/// .expect("Failed to close inotify instance");
/// ```
///
/// [`Inotify`]: struct.Inotify.html
/// [`close`]: ../libc/fn.close.html
pub fn close(self) -> io::Result<()> {
// `self` will be dropped when this method returns. If this is the only
// owner of `fd`, the `Arc` will also be dropped. The `Drop`
// implementation for `FdGuard` will attempt to close the file descriptor
// again, unless this flag here is cleared.
self.fd.should_not_close();
match unsafe { ffi::close(**self.fd) } {
0 => Ok(()),
_ => Err(io::Error::last_os_error()),
}
}
}
impl AsRawFd for Inotify {
#[inline]
fn as_raw_fd(&self) -> RawFd {
self.fd.as_raw_fd()
}
}
impl FromRawFd for Inotify {
unsafe fn from_raw_fd(fd: RawFd) -> Self {
Inotify {
fd: Arc::new(FdGuard::from_raw_fd(fd))
}
}
}
impl IntoRawFd for Inotify {
#[inline]
fn into_raw_fd(self) -> RawFd {
self.fd.should_not_close();
self.fd.fd
}
}