Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Stacked borrows fails on {ChunksMut,ChunksExactMut}::__iterator_get_unchecked() #94231

Closed
andylizi opened this issue Feb 21, 2022 · 6 comments · Fixed by #94247
Closed

Stacked borrows fails on {ChunksMut,ChunksExactMut}::__iterator_get_unchecked() #94231

andylizi opened this issue Feb 21, 2022 · 6 comments · Fixed by #94247
Labels
T-libs Relevant to the library team, which will review and decide on the PR/issue.

Comments

@andylizi
Copy link
Contributor

andylizi commented Feb 21, 2022

Description

fn main() {
    let mut arr1 = [0u8; 64];
    let arr2 = [0u8; 64];
    let mut iter = arr1.chunks_mut(8).zip(arr2.chunks(8));
    while let Some((chunk1, chunk2)) = iter.next() {
        dbg!(chunk2[0]);
        dbg!(chunk1[0]);
        iter.next();
        dbg!(chunk2[0]);
        dbg!(chunk1[0]);  // error here
    }
}

Running this code in Miri will produce the following output:

[src\main.rs:6] chunk2[0] = 0
[src\main.rs:7] chunk1[0] = 0
[src\main.rs:9] chunk2[0] = 0
error: Undefined Behavior: no item granting read access to tag <1832> at alloc906 found in borrow stack.
  --> src\main.rs:10:9
   |
10 |         dbg!(chunk1[0]);  // error here
   |         ^^^^^^^^^^^^^^^ no item granting read access to tag <1832> at alloc906 found in borrow stack.
   |
   = help: this indicates a potential bug in the program: it performed an invalid operation, but the rules it violated are still experimental
   = help: see https://github.com/rust-lang/unsafe-code-guidelines/blob/master/wip/stacked-borrows.md for further information

With -Zmiri-track-pointer-tag=1832,1585,12319

note: tracking was triggered
    --> \library\core\src\slice\iter.rs:1564:19
     |
1564 |         Self { v: slice, chunk_size: size }
     |                   ^^^^^ created tag 1585
     |
     = note: inside `std::slice::ChunksMut::<u8>::new` at \library\core\src\slice\iter.rs:1564:19
     = note: inside `core::slice::<impl [u8]>::chunks_mut` at \library\core\src\slice\mod.rs:828:9
note: inside `main` at src\main.rs:4:20
    --> src\main.rs:4:20
     |
4    |     let mut iter = arr1.chunks_mut(8).zip(arr2.chunks(8));
     |                    ^^^^^^^^^^^^^^^^^^

note: tracking was triggered
 --> src\main.rs:5:21
  |
5 |     while let Some((chunk1, chunk2)) = iter.next() {
  |                     ^^^^^^ created tag 1832
  |
  = note: inside `main` at src\main.rs:5:21

[src\main.rs:6] chunk2[0] = 0
[src\main.rs:7] chunk1[0] = 0

note: tracking was triggered
    --> \library\core\src\slice\iter.rs:1641:32
     |
1641 |             let len = cmp::min(self.v.len().unchecked_sub(start), self.chunk_size);
     |                                ^^^^^^^^^^^^ popped tracked tag for item [Unique for <1832>] due to Read access for <1585>
     |
     = note: inside `<std::slice::ChunksMut<u8> as std::iter::Iterator>::__iterator_get_unchecked` at \library\core\src\slice\iter.rs:1641:32
     = note: inside `<std::iter::Zip<std::slice::ChunksMut<u8>, std::slice::Chunks<u8>> as std::iter::adapters::zip::ZipImpl<std::slice::ChunksMut<u8>, std::slice::Chunks<u8>>>::next` at \library\core\src\iter\adapters\zip.rs:278:23
     = note: inside `<std::iter::Zip<std::slice::ChunksMut<u8>, std::slice::Chunks<u8>> as std::iter::Iterator>::next` at \library\core\src\iter\adapters\zip.rs:84:9
    --> \library\core\src\slice\iter.rs:1642:32
     |
1642 |             from_raw_parts_mut(self.v.as_mut_ptr().add(start), len)
     |                                ^^^^^^^^^^^^^^^^^^^ created tag 12319
     |
     = note: inside `<std::slice::ChunksMut<u8> as std::iter::Iterator>::__iterator_get_unchecked` at \library\core\src\slice\iter.rs:1642:32
    --> \library\core\src\slice\mod.rs:483:5
      |
  483 | /     pub const fn as_mut_ptr(&mut self) -> *mut T {
  484 | |         self as *mut [T] as *mut T
  485 | |     }
      | |_____^ popped tracked tag for item [Disabled for <1832>] due to Write access for <12319>
      |
      = note: inside `core::slice::<impl [u8]>::as_mut_ptr` at \library\core\src\slice\mod.rs:483:5
      = note: inside `<std::slice::ChunksMut<u8> as std::iter::Iterator>::__iterator_get_unchecked` at \library\core\src\slice\iter.rs:1642:32
     note: inside `main` at src\main.rs:8:9
    --> src\main.rs:8:9
     |
8    |         iter.next();
     |         ^^^^^^^^^^^
     = note

[src\main.rs:9] chunk2[0] = 0

error: Undefined Behavior: no item granting read access to tag <1832> at alloc906 found in borrow stack.
  --> src\main.rs:10:9
   |
10 |         dbg!(chunk1[0]);  // error here
   |         ^^^^^^^^^^^^^^^ no item granting read access to tag <1832> at alloc906 found in borrow stack.
   |
   = help: this indicates a potential bug in the program: it performed an invalid operation, but the rules it violated are still experimental
   = help: see https://github.com/rust-lang/unsafe-code-guidelines/blob/master/wip/stacked-borrows.md for further information

A few things to note here:

  1. Access to the mutable chunk will fail. For example arr1.chunks(8).zip(arr2.chunks_mut(8)) will fail on chunk2[0] instead.
  2. zip() is required. This is probably due to Zip uses __iterator_get_unchecked() internally rather than next().
  3. chunks_mut() and chunks_exact_mut() can both reproduce.
  4. array_chunks_mut() can't reproduce.

Environment

$ rustc --version --verbose
rustc 1.61.0-nightly (45e2c2881 2022-02-20)
binary: rustc
commit-hash: 45e2c2881d11324d610815bfff097e25c412199e
commit-date: 2022-02-20
host: x86_64-pc-windows-msvc
release: 1.61.0-nightly
LLVM version: 14.0.0

$ cargo miri --version
miri 0.1.0 (0db4090 2022-02-12)
@andylizi andylizi changed the title Stacked Borrows failed on chunks_mut() + zip() Stacked borrows fails on {ChunksMut,ChunksExactMut}::__iterator_get_unchecked() Feb 21, 2022
@RalfJung
Copy link
Member

There's probably more of these kinds of problems that @saethlin has been looking into recently.

@saethlin
Copy link
Member

saethlin commented Feb 21, 2022

Huge thanks for reporting this @andylizi!

This is definitely an issue with the implementation in core. Perhaps this issue should be moved to rust-lang/rust? (as in, by someone with permission to move the issue itself, not closed and a new one opened)

The existing definition of ChunksMut:

pub struct ChunksMut<'a, T: 'a> {
v: &'a mut [T],
chunk_size: usize,
}

Cannot support the current implementation of __iterator_get_unchecked:
unsafe fn __iterator_get_unchecked(&mut self, idx: usize) -> Self::Item {
let start = idx * self.chunk_size;
// SAFETY: see comments for `Chunks::__iterator_get_unchecked`.
//
// Also note that the caller also guarantees that we're never called
// with the same index again, and that no other methods that will
// access this subslice are called, so it is valid for the returned
// slice to be mutable.
unsafe {
let len = cmp::min(self.v.len().unchecked_sub(start), self.chunk_size);
from_raw_parts_mut(self.v.as_mut_ptr().add(start), len)
}
}

The problem is that this method is trying to hand out subslices of a &mut [T] while also holding on to the original. Because it wraps a &mut [T], ChunksMut guarantees that its inner slice always shrinks when it returns a slice. It looks like this guarantee was accidentally introduced and accidentally upheld in implementation by avoiding unsafe... until __iterator_get_unchecked was added.

I'm working on a patch that fixes the aliasing problem.

It's perhaps worth noting that the specialization on zip that this exists to support has already earned 4 I-unsound issues, and this aliasing issue isn't due to new code. We/I should have caught this already, but didn't. That's quite annoying.

@RalfJung RalfJung transferred this issue from rust-lang/miri Feb 21, 2022
@RalfJung
Copy link
Member

All right, I moved this to the Rust repo.

@the8472
Copy link
Member

the8472 commented Feb 22, 2022

has already earned 4 I-unsound issues

For large values of 4.
#85873, #85863, #82282, #80670, #82291, #81740, #86443

@scottmcm scottmcm added T-libs Relevant to the library team, which will review and decide on the PR/issue. I-libs-nominated Nominated for discussion during a libs team meeting. labels Mar 29, 2022
@scottmcm
Copy link
Member

Nominating for libs; it might be worth a conversation about whether this is proving too unreliable to keep doing in the current way.

@scottmcm scottmcm removed the I-libs-nominated Nominated for discussion during a libs team meeting. label Mar 30, 2022
@the8472
Copy link
Member

the8472 commented Mar 31, 2022

There's an existing test that covers the combination of ChunksMut and Zip (which exercises get_unchecked) that didn't catch this, so unsafe-code-test-coverage (which was discussed in a recent meeting) wouldn't have helped here.

#[test]
fn test_chunks_mut_zip() {
let v1: &mut [i32] = &mut [0, 1, 2, 3, 4];
let v2: &[i32] = &[6, 7, 8, 9, 10];
for (a, b) in v1.chunks_mut(2).zip(v2.chunks(2)) {
let sum = b.iter().sum::<i32>();
for v in a {
*v += sum;
}
}
assert_eq!(v1, [13, 14, 19, 20, 14]);
}

@bors bors closed this as completed in ef81fca Jul 27, 2022
workingjubilee pushed a commit to tcdi/postgrestd that referenced this issue Sep 15, 2022
Fix slice::ChunksMut aliasing

Fixes rust-lang/rust#94231, details in that issue.
cc `@RalfJung`

This isn't done just yet, all the safety comments are placeholders. But otherwise, it seems to work.

I don't really like this approach though. There's a lot of unsafe code where there wasn't before, but as far as I can tell the only other way to uphold the aliasing requirement imposed by `__iterator_get_unchecked` is to use raw slices, which I think require the same amount of unsafe code. All that would do is tie the `len` and `ptr` fields together.

Oh I just looked and I'm pretty sure that `ChunksExactMut`, `RChunksMut`, and `RChunksExactMut` also need to be patched. Even more reason to put up a draft.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
T-libs Relevant to the library team, which will review and decide on the PR/issue.
Projects
None yet
Development

Successfully merging a pull request may close this issue.

5 participants