-
Notifications
You must be signed in to change notification settings - Fork 12.7k
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
Data race in thread::scope #98498
Comments
The following patch fixes this: diff --git a/library/std/src/thread/scoped.rs b/library/std/src/thread/scoped.rs
index a387a09dc8b..06019cbc75b 100644
--- a/library/std/src/thread/scoped.rs
+++ b/library/std/src/thread/scoped.rs
@@ -55,8 +55,10 @@ pub(super) fn decrement_num_running_threads(&self, panic: bool) {
if panic {
self.a_thread_panicked.store(true, Ordering::Relaxed);
}
+ // The moment we do the `fetch_sub`, the memory we ourselves are stored in can be deallocated.
+ let main_thread = self.main_thread.clone();
if self.num_running_threads.fetch_sub(1, Ordering::Release) == 1 {
- self.main_thread.unpark();
+ main_thread.unpark();
}
}
} However, even with this patch, we still run into a problem like #55005: the Another alternative might be to put the entire |
Going to attach this to the 1.63 milestone, since this is in a newly stabilized feature -- we may wish to back out stabilization for a cycle or backport fixes. |
fix data race in thread::scope Puts the `ScopeData` into an `Arc` so it sticks around as long as we need it. This means one extra `Arc::clone` per spawned scoped thread, which I hope is fine. Fixes rust-lang#98498 r? `@m-ou-se`
fix data race in thread::scope Puts the `ScopeData` into an `Arc` so it sticks around as long as we need it. This means one extra `Arc::clone` per spawned scoped thread, which I hope is fine. Fixes rust-lang#98498 r? ``@m-ou-se``
fix data race in thread::scope Puts the `ScopeData` into an `Arc` so it sticks around as long as we need it. This means one extra `Arc::clone` per spawned scoped thread, which I hope is fine. Fixes rust-lang#98498 r? ```@m-ou-se```
fix data race in thread::scope Puts the `ScopeData` into an `Arc` so it sticks around as long as we need it. This means one extra `Arc::clone` per spawned scoped thread, which I hope is fine. Fixes rust-lang#98498 r? ````@m-ou-se````
Re-opening for beta backport. |
fix data race in thread::scope Puts the `ScopeData` into an `Arc` so it sticks around as long as we need it. This means one extra `Arc::clone` per spawned scoped thread, which I hope is fine. Fixes rust-lang#98498 r? `````@m-ou-se`````
The fix was backported in #98949. |
fix data race in thread::scope Puts the `ScopeData` into an `Arc` so it sticks around as long as we need it. This means one extra `Arc::clone` per spawned scoped thread, which I hope is fine. Fixes rust-lang/rust#98498 r? `````@m-ou-se`````
The following code sometimes causes a data race to be reported by Miri:
Miri reports:
The deallocation occurs when
scope
returns and its local variablescope
gets deallocated.The read occurs here (I am fairly sure):
rust/library/std/src/thread/scoped.rs
Lines 58 to 59 in 7098a71
This reads the contents of the field
self.main_thread
after thefetch_sub
. But the moment thefetch_sub
is done, the memory backingself
might be deallocated. That makes the read a potential use-after-free, and a race with the deallocation.Cc @m-ou-se
The text was updated successfully, but these errors were encountered: