Miri reports UB in tests

[Timo]

Haven't seen anyone mention this yet so I thought I'd open an issue here. Running the current tests under miri reports Undefined Behavior in Gc<T>'s Drop impl:

test test_derive_bounds ... error: Undefined Behavior: dereferencing pointer failed: 0x1c56a8 is not a valid pointer
   --> /home/timo/rust-gc/gc/src/lib.rs:134:18
    |
134 |         unsafe { &*clear_root_bit(self.ptr_root.get()).as_ptr() }
    |                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ dereferencing pointer failed: 0x1c56a8 is not a valid pointer
    |
    = help: this indicates a bug in the program: it performed an invalid operation, and caused Undefined Behavior
    = help: see https://doc.rust-lang.org/nightly/reference/behavior-considered-undefined.html for further information
    = note: backtrace:
    = note: inside `gc::Gc::<Thunk<NotTrace>>::inner` at /home/timo/rust-gc/gc/src/lib.rs:134:18
note: inside `<gc::Gc<Thunk<NotTrace>> as std::ops::Drop>::drop` at /home/timo/rust-gc/gc/src/lib.rs:279:17
   --> /home/timo/rust-gc/gc/src/lib.rs:279:17
    |
279 |                 self.inner().unroot_inner();
    |                 ^^^^^^^^^^^^

[Manish Goregaokar]

I don't have time to debug this, but I bet this is mostly because this crate was written a while ago before a lot of this stuff was pinned down and it's doing some pointer stuff in a way that's not ideal.

In particular, we're doing pointer masking, which probably falls afoul of strict provenance.

[Anders Kaseorg]

If that is the issue, it might be easy to fix using https://crates.io/crates/sptr.

[Manish Goregaokar]

My understanding was that strict provenance was something that was optional in miri, however. Unsure if I want to add new deps for it ust yet, might want to wait till those things are in stable.

[Anders Kaseorg]

I was able to reproduce and find a simple fix: #146.

[Timo]

Can confirm that PR did fix this specific error, however miri still reports UB (specifically in GcStates Drop code when sweeping objects)

Error

error: Undefined Behavior: attempting a write access using <210556> at alloc88933[0x38], but that tag does not exist in the borrow stack for this location
   --> /workspaces/rust-gc/gc/src/gc.rs:231:13
    |
231 |             *incoming = node.header.next.take();
    |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    |             |
    |             attempting a write access using <210556> at alloc88933[0x38], but that tag does not exist in the borrow stack for this location
    |             this error occurs as part of an access at alloc88933[0x38..0x48]
    |
    = help: this indicates a potential bug in the program: it performed an invalid operation, but the Stacked Borrows 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
help: <210556> was created by a retag at offsets [0x38..0x48]
   --> /workspaces/rust-gc/gc/src/gc.rs:213:31
    |
213 |                     incoming: unmark_head,
    |                               ^^^^^^^^^^^
help: <210556> was later invalidated at offsets [0x38..0x48]
   --> /workspaces/rust-gc/gc/src/gc.rs:243:14
    |
243 |         mark(&mut st.boxes_start);
    |              ^^^^^^^^^^^^^^^^^^^
    = note: backtrace:
    = note: inside `gc::gc::collect_garbage::sweep` at /workspaces/rust-gc/gc/src/gc.rs:231:13
note: inside `gc::gc::collect_garbage` at /workspaces/rust-gc/gc/src/gc.rs:244:9
   --> /workspaces/rust-gc/gc/src/gc.rs:244:9
    |
244 |         sweep(unmarked, &mut st.stats.bytes_allocated);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
note: inside `<gc::gc::GcState as std::ops::Drop>::drop` at /workspaces/rust-gc/gc/src/gc.rs:15:13
   --> /workspaces/rust-gc/gc/src/gc.rs:15:13
    |
15  |             collect_garbage(self);
    |             ^^^^^^^^^^^^^^^^^^^^^

note: some details are omitted, run with `MIRIFLAGS=-Zmiri-backtrace=full` for a verbose backtrace

Interestingly this error doesn't have an associated test and all tests pass, but I suppose that's just because it happens in the destructor of a threadlocal.

[Anders Kaseorg]

Oh hmm, that one is a new error with this week’s nightlies.

• rustc 1.64.0-nightly (2f3ddd9f5 2022-06-27), miri 0.1.0 (9e2dac4 2022-06-26): works
• rustc 1.64.0-nightly (7425fb293 2022-06-30), miri 0.1.0 (ff62c3a 2022-06-30): fails

Probably due to rust-lang/miri#2275, which was merged within that range. Indeed, earlier nightlies fail in the same way with MIRIFLAGS=-Zmiri-permissive-provenance.

[Anders Kaseorg]

#147 seems to fix everything.