kernel/
iov.rs

1// SPDX-License-Identifier: GPL-2.0
2
3// Copyright (C) 2025 Google LLC.
4
5//! IO vectors.
6//!
7//! C headers: [`include/linux/iov_iter.h`](srctree/include/linux/iov_iter.h),
8//! [`include/linux/uio.h`](srctree/include/linux/uio.h)
9
10use crate::{
11    alloc::{Allocator, Flags},
12    bindings,
13    prelude::*,
14    types::Opaque,
15};
16use core::{marker::PhantomData, mem::MaybeUninit, ptr, slice};
17
18const ITER_SOURCE: bool = bindings::ITER_SOURCE != 0;
19const ITER_DEST: bool = bindings::ITER_DEST != 0;
20
21// Compile-time assertion for the above constants.
22const _: () = {
23    build_assert!(
24        ITER_SOURCE != ITER_DEST,
25        "ITER_DEST and ITER_SOURCE should be different."
26    );
27};
28
29/// An IO vector that acts as a source of data.
30///
31/// The data may come from many different sources. This includes both things in kernel-space and
32/// reading from userspace. It's not necessarily the case that the data source is immutable, so
33/// rewinding the IO vector to read the same data twice is not guaranteed to result in the same
34/// bytes. It's also possible that the data source is mapped in a thread-local manner using e.g.
35/// `kmap_local_page()`, so this type is not `Send` to ensure that the mapping is read from the
36/// right context in that scenario.
37///
38/// # Invariants
39///
40/// Must hold a valid `struct iov_iter` with `data_source` set to `ITER_SOURCE`. For the duration
41/// of `'data`, it must be safe to read from this IO vector using the standard C methods for this
42/// purpose.
43#[repr(transparent)]
44pub struct IovIterSource<'data> {
45    iov: Opaque<bindings::iov_iter>,
46    /// Represent to the type system that this value contains a pointer to readable data it does
47    /// not own.
48    _source: PhantomData<&'data [u8]>,
49}
50
51impl<'data> IovIterSource<'data> {
52    /// Obtain an `IovIterSource` from a raw pointer.
53    ///
54    /// # Safety
55    ///
56    /// * The referenced `struct iov_iter` must be valid and must only be accessed through the
57    ///   returned reference for the duration of `'iov`.
58    /// * The referenced `struct iov_iter` must have `data_source` set to `ITER_SOURCE`.
59    /// * For the duration of `'data`, it must be safe to read from this IO vector using the
60    ///   standard C methods for this purpose.
61    #[track_caller]
62    #[inline]
63    pub unsafe fn from_raw<'iov>(ptr: *mut bindings::iov_iter) -> &'iov mut IovIterSource<'data> {
64        // SAFETY: The caller ensures that `ptr` is valid.
65        let data_source = unsafe { (*ptr).data_source };
66        assert_eq!(data_source, ITER_SOURCE);
67
68        // SAFETY: The caller ensures the type invariants for the right durations, and
69        // `IovIterSource` is layout compatible with `struct iov_iter`.
70        unsafe { &mut *ptr.cast::<IovIterSource<'data>>() }
71    }
72
73    /// Access this as a raw `struct iov_iter`.
74    #[inline]
75    pub fn as_raw(&mut self) -> *mut bindings::iov_iter {
76        self.iov.get()
77    }
78
79    /// Returns the number of bytes available in this IO vector.
80    ///
81    /// Note that this may overestimate the number of bytes. For example, reading from userspace
82    /// memory could fail with `EFAULT`, which will be treated as the end of the IO vector.
83    #[inline]
84    pub fn len(&self) -> usize {
85        // SAFETY: We have shared access to this IO vector, so we can read its `count` field.
86        unsafe {
87            (*self.iov.get())
88                .__bindgen_anon_1
89                .__bindgen_anon_1
90                .as_ref()
91                .count
92        }
93    }
94
95    /// Returns whether there are any bytes left in this IO vector.
96    ///
97    /// This may return `true` even if there are no more bytes available. For example, reading from
98    /// userspace memory could fail with `EFAULT`, which will be treated as the end of the IO vector.
99    #[inline]
100    pub fn is_empty(&self) -> bool {
101        self.len() == 0
102    }
103
104    /// Advance this IO vector by `bytes` bytes.
105    ///
106    /// If `bytes` is larger than the size of this IO vector, it is advanced to the end.
107    #[inline]
108    pub fn advance(&mut self, bytes: usize) {
109        // SAFETY: By the type invariants, `self.iov` is a valid IO vector.
110        unsafe { bindings::iov_iter_advance(self.as_raw(), bytes) };
111    }
112
113    /// Advance this IO vector backwards by `bytes` bytes.
114    ///
115    /// # Safety
116    ///
117    /// The IO vector must not be reverted to before its beginning.
118    #[inline]
119    pub unsafe fn revert(&mut self, bytes: usize) {
120        // SAFETY: By the type invariants, `self.iov` is a valid IO vector, and the caller
121        // ensures that `bytes` is in bounds.
122        unsafe { bindings::iov_iter_revert(self.as_raw(), bytes) };
123    }
124
125    /// Read data from this IO vector.
126    ///
127    /// Returns the number of bytes that have been copied.
128    #[inline]
129    pub fn copy_from_iter(&mut self, out: &mut [u8]) -> usize {
130        // SAFETY: `Self::copy_from_iter_raw` guarantees that it will not write any uninitialized
131        // bytes in the provided buffer, so `out` is still a valid `u8` slice after this call.
132        let out = unsafe { &mut *(ptr::from_mut(out) as *mut [MaybeUninit<u8>]) };
133
134        self.copy_from_iter_raw(out).len()
135    }
136
137    /// Read data from this IO vector and append it to a vector.
138    ///
139    /// Returns the number of bytes that have been copied.
140    #[inline]
141    pub fn copy_from_iter_vec<A: Allocator>(
142        &mut self,
143        out: &mut Vec<u8, A>,
144        flags: Flags,
145    ) -> Result<usize> {
146        out.reserve(self.len(), flags)?;
147        let len = self.copy_from_iter_raw(out.spare_capacity_mut()).len();
148        // SAFETY:
149        // - `len` is the length of a subslice of the spare capacity, so `len` is at most the
150        //   length of the spare capacity.
151        // - `Self::copy_from_iter_raw` guarantees that the first `len` bytes of the spare capacity
152        //   have been initialized.
153        unsafe { out.inc_len(len) };
154        Ok(len)
155    }
156
157    /// Read data from this IO vector into potentially uninitialized memory.
158    ///
159    /// Returns the sub-slice of the output that has been initialized. If the returned slice is
160    /// shorter than the input buffer, then the entire IO vector has been read.
161    ///
162    /// This will never write uninitialized bytes to the provided buffer.
163    #[inline]
164    pub fn copy_from_iter_raw(&mut self, out: &mut [MaybeUninit<u8>]) -> &mut [u8] {
165        let capacity = out.len();
166        let out = out.as_mut_ptr().cast::<u8>();
167
168        // GUARANTEES: The C API guarantees that it does not write uninitialized bytes to the
169        // provided buffer.
170        // SAFETY:
171        // * By the type invariants, it is still valid to read from this IO vector.
172        // * `out` is valid for writing for `capacity` bytes because it comes from a slice of
173        //   that length.
174        let len = unsafe { bindings::_copy_from_iter(out.cast(), capacity, self.as_raw()) };
175
176        // SAFETY: The underlying C api guarantees that initialized bytes have been written to the
177        // first `len` bytes of the spare capacity.
178        unsafe { slice::from_raw_parts_mut(out, len) }
179    }
180}
181
182/// An IO vector that acts as a destination for data.
183///
184/// IO vectors support many different types of destinations. This includes both buffers in
185/// kernel-space and writing to userspace. It's possible that the destination buffer is mapped in a
186/// thread-local manner using e.g. `kmap_local_page()`, so this type is not `Send` to ensure that
187/// the mapping is written to the right context in that scenario.
188///
189/// # Invariants
190///
191/// Must hold a valid `struct iov_iter` with `data_source` set to `ITER_DEST`. For the duration of
192/// `'data`, it must be safe to write to this IO vector using the standard C methods for this
193/// purpose.
194#[repr(transparent)]
195pub struct IovIterDest<'data> {
196    iov: Opaque<bindings::iov_iter>,
197    /// Represent to the type system that this value contains a pointer to writable data it does
198    /// not own.
199    _source: PhantomData<&'data mut [u8]>,
200}
201
202impl<'data> IovIterDest<'data> {
203    /// Obtain an `IovIterDest` from a raw pointer.
204    ///
205    /// # Safety
206    ///
207    /// * The referenced `struct iov_iter` must be valid and must only be accessed through the
208    ///   returned reference for the duration of `'iov`.
209    /// * The referenced `struct iov_iter` must have `data_source` set to `ITER_DEST`.
210    /// * For the duration of `'data`, it must be safe to write to this IO vector using the
211    ///   standard C methods for this purpose.
212    #[track_caller]
213    #[inline]
214    pub unsafe fn from_raw<'iov>(ptr: *mut bindings::iov_iter) -> &'iov mut IovIterDest<'data> {
215        // SAFETY: The caller ensures that `ptr` is valid.
216        let data_source = unsafe { (*ptr).data_source };
217        assert_eq!(data_source, ITER_DEST);
218
219        // SAFETY: The caller ensures the type invariants for the right durations, and
220        // `IovIterSource` is layout compatible with `struct iov_iter`.
221        unsafe { &mut *ptr.cast::<IovIterDest<'data>>() }
222    }
223
224    /// Access this as a raw `struct iov_iter`.
225    #[inline]
226    pub fn as_raw(&mut self) -> *mut bindings::iov_iter {
227        self.iov.get()
228    }
229
230    /// Returns the number of bytes available in this IO vector.
231    ///
232    /// Note that this may overestimate the number of bytes. For example, reading from userspace
233    /// memory could fail with EFAULT, which will be treated as the end of the IO vector.
234    #[inline]
235    pub fn len(&self) -> usize {
236        // SAFETY: We have shared access to this IO vector, so we can read its `count` field.
237        unsafe {
238            (*self.iov.get())
239                .__bindgen_anon_1
240                .__bindgen_anon_1
241                .as_ref()
242                .count
243        }
244    }
245
246    /// Returns whether there are any bytes left in this IO vector.
247    ///
248    /// This may return `true` even if there are no more bytes available. For example, reading from
249    /// userspace memory could fail with EFAULT, which will be treated as the end of the IO vector.
250    #[inline]
251    pub fn is_empty(&self) -> bool {
252        self.len() == 0
253    }
254
255    /// Advance this IO vector by `bytes` bytes.
256    ///
257    /// If `bytes` is larger than the size of this IO vector, it is advanced to the end.
258    #[inline]
259    pub fn advance(&mut self, bytes: usize) {
260        // SAFETY: By the type invariants, `self.iov` is a valid IO vector.
261        unsafe { bindings::iov_iter_advance(self.as_raw(), bytes) };
262    }
263
264    /// Advance this IO vector backwards by `bytes` bytes.
265    ///
266    /// # Safety
267    ///
268    /// The IO vector must not be reverted to before its beginning.
269    #[inline]
270    pub unsafe fn revert(&mut self, bytes: usize) {
271        // SAFETY: By the type invariants, `self.iov` is a valid IO vector, and the caller
272        // ensures that `bytes` is in bounds.
273        unsafe { bindings::iov_iter_revert(self.as_raw(), bytes) };
274    }
275
276    /// Write data to this IO vector.
277    ///
278    /// Returns the number of bytes that were written. If this is shorter than the provided slice,
279    /// then no more bytes can be written.
280    #[inline]
281    pub fn copy_to_iter(&mut self, input: &[u8]) -> usize {
282        // SAFETY:
283        // * By the type invariants, it is still valid to write to this IO vector.
284        // * `input` is valid for `input.len()` bytes.
285        unsafe { bindings::_copy_to_iter(input.as_ptr().cast(), input.len(), self.as_raw()) }
286    }
287
288    /// Utility for implementing `read_iter` given the full contents of the file.
289    ///
290    /// The full contents of the file being read from is represented by `contents`. This call will
291    /// write the appropriate sub-slice of `contents` and update the file position in `ppos` so
292    /// that the file will appear to contain `contents` even if takes multiple reads to read the
293    /// entire file.
294    #[inline]
295    pub fn simple_read_from_buffer(&mut self, ppos: &mut i64, contents: &[u8]) -> Result<usize> {
296        if *ppos < 0 {
297            return Err(EINVAL);
298        }
299        let Ok(pos) = usize::try_from(*ppos) else {
300            return Ok(0);
301        };
302        if pos >= contents.len() {
303            return Ok(0);
304        }
305
306        // BOUNDS: We just checked that `pos < contents.len()` above.
307        let num_written = self.copy_to_iter(&contents[pos..]);
308
309        // OVERFLOW: `pos+num_written <= contents.len() <= isize::MAX <= i64::MAX`.
310        *ppos = (pos + num_written) as i64;
311
312        Ok(num_written)
313    }
314}