Source code

Revision control

Copy as Markdown

Other Tools

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
use crate::ffi::{rust_call, ForeignBytes, RustCallStatus};
/// Support for passing an allocated-by-Rust buffer of bytes over the FFI.
///
/// We can pass a `Vec<u8>` to foreign language code by decomposing it into
/// its raw parts (buffer pointer, length, and capacity) and passing those
/// around as a struct. Naturally, this can be tremendously unsafe! So here
/// are the details:
///
/// * `RustBuffer` structs must only ever be constructed from a `Vec<u8>`,
/// either explicitly via `RustBuffer::from_vec` or indirectly by calling
/// one of the `RustBuffer::new*` constructors.
///
/// * `RustBuffer` structs do not implement `Drop`, since they are intended
/// to be passed to foreign-language code outside of the control of Rust's
/// ownership system. To avoid memory leaks they *must* passed back into
/// Rust and either explicitly destroyed using `RustBuffer::destroy`, or
/// converted back to a `Vec<u8>` using `RustBuffer::destroy_into_vec`
/// (which will then be dropped via Rust's usual ownership-tracking system).
///
/// Foreign-language code should not construct `RustBuffer` structs other than
/// by receiving them from a call into the Rust code, and should not modify them
/// apart from the following safe operations:
///
/// * Writing bytes into the buffer pointed to by `data`, without writing
/// beyond the indicated `capacity`.
///
/// * Adjusting the `len` property to indicate the amount of data written,
/// while ensuring that 0 <= `len` <= `capacity`.
///
/// * As a special case, constructing a `RustBuffer` with zero capacity, zero
/// length, and a null `data` pointer to indicate an empty buffer.
///
/// In particular, it is not safe for foreign-language code to construct a `RustBuffer`
/// that points to its own allocated memory; use the `ForeignBytes` struct to
/// pass a view of foreign-owned memory in to Rust code.
///
/// Implementation note: all the fields of this struct are private, so you can't
/// manually construct instances that don't come from a `Vec<u8>`. If you've got
/// a `RustBuffer` then it either came from a public constructor (all of which
/// are safe) or it came from foreign-language code (which should have in turn
/// received it by calling some Rust function, and should be respecting the
/// invariants listed above).
///
/// This struct is based on `ByteBuffer` from the `ffi-support` crate, but modified
/// to retain unallocated capacity rather than truncating to the occupied length.
#[repr(C)]
#[derive(Debug)]
pub struct RustBuffer {
/// The allocated capacity of the underlying `Vec<u8>`.
/// In Rust this is a `usize`, but we use an `u64` to keep the foreign binding code simple.
pub(crate) capacity: u64,
/// The occupied length of the underlying `Vec<u8>`.
/// In Rust this is a `usize`, but we use an `u64` to keep the foreign binding code simple.
pub(crate) len: u64,
/// The pointer to the allocated buffer of the `Vec<u8>`.
pub(crate) data: *mut u8,
}
// Mark `RustBuffer` as safe to send between threads, despite the `u8` pointer. The only mutable
// use of that pointer is in `destroy_into_vec()` which requires a &mut on the `RustBuffer`. This
// is required to send `RustBuffer` inside a `RustFuture`
unsafe impl Send for RustBuffer {}
impl RustBuffer {
/// Creates an empty `RustBuffer`.
///
/// The buffer will not allocate.
/// The resulting vector will not be automatically dropped; you must
/// arrange to call `destroy` or `destroy_into_vec` when finished with it.
pub fn new() -> Self {
Self::from_vec(Vec::new())
}
/// Creates a `RustBuffer` from its constituent fields.
///
/// # Safety
///
/// You must ensure that the raw parts uphold the documented invariants of this class.
pub(crate) unsafe fn from_raw_parts(data: *mut u8, len: u64, capacity: u64) -> Self {
Self {
capacity,
len,
data,
}
}
/// Get the current length of the buffer, as a `usize`.
///
/// This is mostly a helper function to convert the `u64` length field
/// into a `usize`, which is what Rust code usually expects.
///
/// # Panics
///
/// Panics if called on an invalid struct obtained from foreign-language code,
/// in which the `len` field is negative.
pub fn len(&self) -> usize {
self.len
.try_into()
.expect("buffer length negative or overflowed")
}
pub fn capacity(&self) -> usize {
self.capacity
.try_into()
.expect("buffer length negative or overflowed")
}
/// Get a pointer to the data
pub fn data_pointer(&self) -> *const u8 {
self.data
}
/// Returns true if the length of the buffer is 0.
pub fn is_empty(&self) -> bool {
self.len == 0
}
/// Creates a `RustBuffer` zero-filed to the requested size.
///
/// The resulting vector will not be automatically dropped; you must
/// arrange to call `destroy` or `destroy_into_vec` when finished with it.
///
/// # Panics
///
/// Panics if the requested size is too large to fit in an `u64`, and
/// hence would risk incompatibility with some foreign-language code.
pub fn new_with_size(size: u64) -> Self {
Self::from_vec(vec![0u8; size as usize])
}
/// Consumes a `Vec<u8>` and returns its raw parts as a `RustBuffer`.
///
/// The resulting vector will not be automatically dropped; you must
/// arrange to call `destroy` or `destroy_into_vec` when finished with it.
///
/// # Panics
///
/// Panics if the vector's length or capacity are too large to fit in an `u64`,
/// and hence would risk incompatibility with some foreign-language code.
pub fn from_vec(v: Vec<u8>) -> Self {
let capacity = u64::try_from(v.capacity()).expect("buffer capacity cannot fit into a u64.");
let len = u64::try_from(v.len()).expect("buffer length cannot fit into a u64.");
let mut v = std::mem::ManuallyDrop::new(v);
unsafe { Self::from_raw_parts(v.as_mut_ptr(), len, capacity) }
}
/// Converts this `RustBuffer` back into an owned `Vec<u8>`.
///
/// This restores ownership of the underlying buffer to Rust, meaning it will
/// be dropped when the `Vec<u8>` is dropped. The `RustBuffer` *must* have been
/// previously obtained from a valid `Vec<u8>` owned by this Rust code.
///
/// # Panics
///
/// Panics if called on an invalid struct obtained from foreign-language code,
/// which does not respect the invairiants on `len` and `capacity`.
pub fn destroy_into_vec(self) -> Vec<u8> {
// Rust will never give us a null `data` pointer for a `Vec`, but
// foreign-language code can use it to cheaply pass an empty buffer.
if self.data.is_null() {
assert!(self.capacity == 0, "null RustBuffer had non-zero capacity");
assert!(self.len == 0, "null RustBuffer had non-zero length");
vec![]
} else {
let capacity: usize = self
.capacity
.try_into()
.expect("buffer capacity negative or overflowed");
let len: usize = self
.len
.try_into()
.expect("buffer length negative or overflowed");
assert!(len <= capacity, "RustBuffer length exceeds capacity");
unsafe { Vec::from_raw_parts(self.data, len, capacity) }
}
}
/// Reclaim memory stored in this `RustBuffer`.
///
/// # Panics
///
/// Panics if called on an invalid struct obtained from foreign-language code,
/// which does not respect the invairiants on `len` and `capacity`.
pub fn destroy(self) {
drop(self.destroy_into_vec());
}
}
impl Default for RustBuffer {
fn default() -> Self {
Self::new()
}
}
// Functions for the RustBuffer functionality.
//
// The scaffolding code re-exports these functions, prefixed with the component name and UDL hash
// This creates a separate set of functions for each UniFFIed component, which is needed in the
// case where we create multiple dylib artifacts since each dylib will have its own allocator.
/// This helper allocates a new byte buffer owned by the Rust code, and returns it
/// to the foreign-language code as a `RustBuffer` struct. Callers must eventually
/// free the resulting buffer, either by explicitly calling [`uniffi_rustbuffer_free`] defined
/// below, or by passing ownership of the buffer back into Rust code.
pub fn uniffi_rustbuffer_alloc(size: u64, call_status: &mut RustCallStatus) -> RustBuffer {
rust_call(call_status, || Ok(RustBuffer::new_with_size(size)))
}
/// This helper copies bytes owned by the foreign-language code into a new byte buffer owned
/// by the Rust code, and returns it as a `RustBuffer` struct. Callers must eventually
/// free the resulting buffer, either by explicitly calling the destructor defined below,
/// or by passing ownership of the buffer back into Rust code.
///
/// # Safety
/// This function will dereference a provided pointer in order to copy bytes from it, so
/// make sure the `ForeignBytes` struct contains a valid pointer and length.
pub fn uniffi_rustbuffer_from_bytes(
bytes: ForeignBytes,
call_status: &mut RustCallStatus,
) -> RustBuffer {
rust_call(call_status, || {
let bytes = bytes.as_slice();
Ok(RustBuffer::from_vec(bytes.to_vec()))
})
}
/// Free a byte buffer that had previously been passed to the foreign language code.
///
/// # Safety
/// The argument *must* be a uniquely-owned `RustBuffer` previously obtained from a call
/// into the Rust code that returned a buffer, or you'll risk freeing unowned memory or
/// corrupting the allocator state.
pub fn uniffi_rustbuffer_free(buf: RustBuffer, call_status: &mut RustCallStatus) {
rust_call(call_status, || {
RustBuffer::destroy(buf);
Ok(())
})
}
/// Reserve additional capacity in a byte buffer that had previously been passed to the
/// foreign language code.
///
/// The first argument *must* be a uniquely-owned `RustBuffer` previously
/// obtained from a call into the Rust code that returned a buffer. Its underlying data pointer
/// will be reallocated if necessary and returned in a new `RustBuffer` struct.
///
/// The second argument must be the minimum number of *additional* bytes to reserve
/// capacity for in the buffer; it is likely to reserve additional capacity in practice
/// due to amortized growth strategy of Rust vectors.
///
/// # Safety
/// The first argument *must* be a uniquely-owned `RustBuffer` previously obtained from a call
/// into the Rust code that returned a buffer, or you'll risk freeing unowned memory or
/// corrupting the allocator state.
pub fn uniffi_rustbuffer_reserve(
buf: RustBuffer,
additional: u64,
call_status: &mut RustCallStatus,
) -> RustBuffer {
rust_call(call_status, || {
let additional: usize = additional
.try_into()
.expect("additional buffer length negative or overflowed");
let mut v = buf.destroy_into_vec();
v.reserve(additional);
Ok(RustBuffer::from_vec(v))
})
}
#[cfg(test)]
mod test {
use super::*;
#[test]
fn test_rustbuffer_from_vec() {
let rbuf = RustBuffer::from_vec(vec![1u8, 2, 3]);
assert_eq!(rbuf.len(), 3);
assert_eq!(rbuf.destroy_into_vec(), vec![1u8, 2, 3]);
}
#[test]
fn test_rustbuffer_empty() {
let rbuf = RustBuffer::new();
assert_eq!(rbuf.len(), 0);
// Rust will never give us a null pointer, even for an empty buffer.
assert!(!rbuf.data.is_null());
assert_eq!(rbuf.destroy_into_vec(), Vec::<u8>::new());
}
#[test]
fn test_rustbuffer_new_with_size() {
let rbuf = RustBuffer::new_with_size(5);
assert_eq!(rbuf.destroy_into_vec().as_slice(), &[0u8, 0, 0, 0, 0]);
let rbuf = RustBuffer::new_with_size(0);
assert!(!rbuf.data.is_null());
assert_eq!(rbuf.destroy_into_vec().as_slice(), &[0u8; 0]);
}
#[test]
fn test_rustbuffer_null_means_empty() {
// This is how foreign-language code might cheaply indicate an empty buffer.
let rbuf = unsafe { RustBuffer::from_raw_parts(std::ptr::null_mut(), 0, 0) };
assert_eq!(rbuf.destroy_into_vec().as_slice(), &[0u8; 0]);
}
#[test]
#[should_panic]
fn test_rustbuffer_null_must_have_no_capacity() {
// We guard against foreign-language code providing this kind of invalid struct.
let rbuf = unsafe { RustBuffer::from_raw_parts(std::ptr::null_mut(), 0, 1) };
rbuf.destroy_into_vec();
}
#[test]
#[should_panic]
fn test_rustbuffer_null_must_have_zero_length() {
// We guard against foreign-language code providing this kind of invalid struct.
let rbuf = unsafe { RustBuffer::from_raw_parts(std::ptr::null_mut(), 12, 0) };
rbuf.destroy_into_vec();
}
#[test]
#[should_panic]
fn test_rustbuffer_provided_len_must_not_exceed_capacity() {
// We guard against foreign-language code providing this kind of invalid struct.
let mut v = vec![0u8, 1, 2];
let rbuf = unsafe { RustBuffer::from_raw_parts(v.as_mut_ptr(), 3, 2) };
rbuf.destroy_into_vec();
}
}