blob: 88473f38fa33d28f30e06370256fc0a7de9180e9 [file] [log] [blame] [edit]
use std::{
ffi::{
OsStr,
OsString,
},
mem,
os::unix::ffi::OsStrExt,
sync::Weak,
};
use inotify_sys as ffi;
use crate::fd_guard::FdGuard;
use crate::watches::WatchDescriptor;
/// Iterator over inotify events
///
/// Allows for iteration over the events returned by
/// [`Inotify::read_events_blocking`] or [`Inotify::read_events`].
///
/// [`Inotify::read_events_blocking`]: crate::Inotify::read_events_blocking
/// [`Inotify::read_events`]: crate::Inotify::read_events
#[derive(Debug)]
pub struct Events<'a> {
fd : Weak<FdGuard>,
buffer : &'a [u8],
num_bytes: usize,
pos : usize,
}
impl<'a> Events<'a> {
pub(crate) fn new(fd: Weak<FdGuard>, buffer: &'a [u8], num_bytes: usize)
-> Self
{
Events {
fd,
buffer,
num_bytes,
pos: 0,
}
}
}
impl<'a> Iterator for Events<'a> {
type Item = Event<&'a OsStr>;
fn next(&mut self) -> Option<Self::Item> {
if self.pos < self.num_bytes {
let (step, event) = Event::from_buffer(self.fd.clone(), &self.buffer[self.pos..]);
self.pos += step;
Some(event)
}
else {
None
}
}
}
/// An inotify event
///
/// A file system event that describes a change that the user previously
/// registered interest in. To watch for events, call [`Watches::add`]. To
/// retrieve events, call [`Inotify::read_events_blocking`] or
/// [`Inotify::read_events`].
///
/// [`Watches::add`]: crate::Watches::add
/// [`Inotify::read_events_blocking`]: crate::Inotify::read_events_blocking
/// [`Inotify::read_events`]: crate::Inotify::read_events
#[derive(Clone, Debug)]
pub struct Event<S> {
/// Identifies the watch this event originates from
///
/// This [`WatchDescriptor`] is equal to the one that [`Watches::add`]
/// returned when interest for this event was registered. The
/// [`WatchDescriptor`] can be used to remove the watch using
/// [`Watches::remove`], thereby preventing future events of this type
/// from being created.
///
/// [`Watches::add`]: crate::Watches::add
/// [`Watches::remove`]: crate::Watches::remove
pub wd: WatchDescriptor,
/// Indicates what kind of event this is
pub mask: EventMask,
/// Connects related events to each other
///
/// When a file is renamed, this results two events: [`MOVED_FROM`] and
/// [`MOVED_TO`]. The `cookie` field will be the same for both of them,
/// thereby making is possible to connect the event pair.
///
/// [`MOVED_FROM`]: EventMask::MOVED_FROM
/// [`MOVED_TO`]: EventMask::MOVED_TO
pub cookie: u32,
/// The name of the file the event originates from
///
/// This field is set only if the subject of the event is a file or directory in a
/// watched directory. If the event concerns a file or directory that is
/// watched directly, `name` will be `None`.
pub name: Option<S>,
}
impl<'a> Event<&'a OsStr> {
fn new(fd: Weak<FdGuard>, event: &ffi::inotify_event, name: &'a OsStr)
-> Self
{
let mask = EventMask::from_bits(event.mask)
.expect("Failed to convert event mask. This indicates a bug.");
let wd = crate::WatchDescriptor {
id: event.wd,
fd,
};
let name = if name.is_empty() {
None
}
else {
Some(name)
};
Event {
wd,
mask,
cookie: event.cookie,
name,
}
}
/// Create an `Event` from a buffer
///
/// Assumes that a full `inotify_event` plus its name is located at the
/// beginning of `buffer`.
///
/// Returns the number of bytes used from the buffer, and the event.
///
/// # Panics
///
/// Panics if the buffer does not contain a full event, including its name.
pub(crate) fn from_buffer(
fd : Weak<FdGuard>,
buffer: &'a [u8],
)
-> (usize, Self)
{
let event_size = mem::size_of::<ffi::inotify_event>();
// Make sure that the buffer is big enough to contain an event, without
// the name. Otherwise we can't safely convert it to an `inotify_event`.
assert!(buffer.len() >= event_size);
let ffi_event_ptr = buffer.as_ptr() as *const ffi::inotify_event;
// We have a pointer to an `inotify_event`, pointing to the beginning of
// `buffer`. Since we know, as per the assertion above, that there are
// enough bytes in the buffer for at least one event, we can safely
// read that `inotify_event`.
// We call `read_unaligned()` since the byte buffer has alignment 1
// and `inotify_event` has a higher alignment, so `*` cannot be used to dereference
// the unaligned pointer (undefined behavior).
let ffi_event = unsafe { ffi_event_ptr.read_unaligned() };
// The name's length is given by `event.len`. There should always be
// enough bytes left in the buffer to fit the name. Let's make sure that
// is the case.
let bytes_left_in_buffer = buffer.len() - event_size;
assert!(bytes_left_in_buffer >= ffi_event.len as usize);
// Directly after the event struct should be a name, if there's one
// associated with the event. Let's make a new slice that starts with
// that name. If there's no name, this slice might have a length of `0`.
let bytes_consumed = event_size + ffi_event.len as usize;
let name = &buffer[event_size..bytes_consumed];
// Remove trailing '\0' bytes
//
// The events in the buffer are aligned, and `name` is filled up
// with '\0' up to the alignment boundary. Here we remove those
// additional bytes.
//
// The `unwrap` here is safe, because `splitn` always returns at
// least one result, even if the original slice contains no '\0'.
let name = name
.splitn(2, |b| b == &0u8)
.next()
.unwrap();
let event = Event::new(
fd,
&ffi_event,
OsStr::from_bytes(name),
);
(bytes_consumed, event)
}
/// Returns an owned copy of the event.
#[deprecated = "use `to_owned()` instead; methods named `into_owned()` usually take self by value"]
#[allow(clippy::wrong_self_convention)]
pub fn into_owned(&self) -> EventOwned {
self.to_owned()
}
/// Returns an owned copy of the event.
#[must_use = "cloning is often expensive and is not expected to have side effects"]
pub fn to_owned(&self) -> EventOwned {
Event {
wd: self.wd.clone(),
mask: self.mask,
cookie: self.cookie,
name: self.name.map(OsStr::to_os_string),
}
}
}
/// An owned version of `Event`
pub type EventOwned = Event<OsString>;
bitflags! {
/// Indicates the type of an event
///
/// This struct can be retrieved from an [`Event`] via its `mask` field.
/// You can determine the [`Event`]'s type by comparing the `EventMask` to
/// its associated constants.
///
/// Please refer to the documentation of [`Event`] for a usage example.
#[derive(PartialEq, Eq, PartialOrd, Ord, Hash, Debug, Clone, Copy)]
pub struct EventMask: u32 {
/// File was accessed
///
/// When watching a directory, this event is only triggered for objects
/// inside the directory, not the directory itself.
///
/// See [`inotify_sys::IN_ACCESS`].
const ACCESS = ffi::IN_ACCESS;
/// Metadata (permissions, timestamps, ...) changed
///
/// When watching a directory, this event can be triggered for the
/// directory itself, as well as objects inside the directory.
///
/// See [`inotify_sys::IN_ATTRIB`].
const ATTRIB = ffi::IN_ATTRIB;
/// File opened for writing was closed
///
/// When watching a directory, this event is only triggered for objects
/// inside the directory, not the directory itself.
///
/// See [`inotify_sys::IN_CLOSE_WRITE`].
const CLOSE_WRITE = ffi::IN_CLOSE_WRITE;
/// File or directory not opened for writing was closed
///
/// When watching a directory, this event can be triggered for the
/// directory itself, as well as objects inside the directory.
///
/// See [`inotify_sys::IN_CLOSE_NOWRITE`].
const CLOSE_NOWRITE = ffi::IN_CLOSE_NOWRITE;
/// File/directory created in watched directory
///
/// When watching a directory, this event is only triggered for objects
/// inside the directory, not the directory itself.
///
/// See [`inotify_sys::IN_CREATE`].
const CREATE = ffi::IN_CREATE;
/// File/directory deleted from watched directory
///
/// When watching a directory, this event is only triggered for objects
/// inside the directory, not the directory itself.
const DELETE = ffi::IN_DELETE;
/// Watched file/directory was deleted
///
/// See [`inotify_sys::IN_DELETE_SELF`].
const DELETE_SELF = ffi::IN_DELETE_SELF;
/// File was modified
///
/// When watching a directory, this event is only triggered for objects
/// inside the directory, not the directory itself.
///
/// See [`inotify_sys::IN_MODIFY`].
const MODIFY = ffi::IN_MODIFY;
/// Watched file/directory was moved
///
/// See [`inotify_sys::IN_MOVE_SELF`].
const MOVE_SELF = ffi::IN_MOVE_SELF;
/// File was renamed/moved; watched directory contained old name
///
/// When watching a directory, this event is only triggered for objects
/// inside the directory, not the directory itself.
///
/// See [`inotify_sys::IN_MOVED_FROM`].
const MOVED_FROM = ffi::IN_MOVED_FROM;
/// File was renamed/moved; watched directory contains new name
///
/// When watching a directory, this event is only triggered for objects
/// inside the directory, not the directory itself.
///
/// See [`inotify_sys::IN_MOVED_TO`].
const MOVED_TO = ffi::IN_MOVED_TO;
/// File or directory was opened
///
/// When watching a directory, this event can be triggered for the
/// directory itself, as well as objects inside the directory.
///
/// See [`inotify_sys::IN_OPEN`].
const OPEN = ffi::IN_OPEN;
/// Watch was removed
///
/// This event will be generated, if the watch was removed explicitly
/// (via [`Watches::remove`]), or automatically (because the file was
/// deleted or the file system was unmounted).
///
/// See [`inotify_sys::IN_IGNORED`].
///
/// [`Watches::remove`]: crate::Watches::remove
const IGNORED = ffi::IN_IGNORED;
/// Event related to a directory
///
/// The subject of the event is a directory.
///
/// See [`inotify_sys::IN_ISDIR`].
const ISDIR = ffi::IN_ISDIR;
/// Event queue overflowed
///
/// The event queue has overflowed and events have presumably been lost.
///
/// See [`inotify_sys::IN_Q_OVERFLOW`].
const Q_OVERFLOW = ffi::IN_Q_OVERFLOW;
/// File system containing watched object was unmounted.
/// File system was unmounted
///
/// The file system that contained the watched object has been
/// unmounted. An event with [`EventMask::IGNORED`] will subsequently be
/// generated for the same watch descriptor.
///
/// See [`inotify_sys::IN_UNMOUNT`].
const UNMOUNT = ffi::IN_UNMOUNT;
}
}
impl EventMask {
/// Wrapper around [`Self::from_bits_retain`] for backwards compatibility
///
/// # Safety
///
/// This function is not actually unsafe. It is just a wrapper around the
/// safe [`Self::from_bits_retain`].
#[deprecated = "Use the safe `from_bits_retain` method instead"]
pub unsafe fn from_bits_unchecked(bits: u32) -> Self {
Self::from_bits_retain(bits)
}
}
#[cfg(test)]
mod tests {
use std::{
io::prelude::*,
mem,
slice,
sync,
};
use inotify_sys as ffi;
use super::Event;
#[test]
fn from_buffer_should_not_mistake_next_event_for_name_of_previous_event() {
let mut buffer = [0u8; 1024];
// First, put a normal event into the buffer
let event = ffi::inotify_event {
wd: 0,
mask: 0,
cookie: 0,
len: 0, // no name following after event
};
let event = unsafe {
slice::from_raw_parts(
&event as *const _ as *const u8,
mem::size_of_val(&event),
)
};
(&mut buffer[..]).write(event)
.expect("Failed to write into buffer");
// After that event, simulate an event that starts with a non-zero byte.
buffer[mem::size_of_val(&event)] = 1;
// Now create the event and verify that the name is actually `None`, as
// dictated by the value `len` above.
let (_, event) = Event::from_buffer(
sync::Weak::new(),
&buffer,
);
assert_eq!(event.name, None);
}
}