From: Stefan Krah <skrah@bytereef.org>
Date: Tue, 3 Mar 2020 08:19:58 +0000 (+0100)
Subject:  bpo-39776: Lock ++interp->tstate_next_unique_id (GH-18746)
X-Git-Tag: v3.7.7rc1~4
X-Git-Url: http://git.ipfire.org/gitweb.cgi?a=commitdiff_plain;h=852aee69f49c654a03ad1f64d90a78ba8848e2c6;p=thirdparty%2FPython%2Fcpython.git

 bpo-39776: Lock ++interp->tstate_next_unique_id (GH-18746)

- Threads created by PyGILState_Ensure() could have a duplicate tstate->id.

(cherry picked from commit b3b9ade4a3d3fe00d933bcd8fc5c5c755d1024f9)
---

diff --git a/Misc/NEWS.d/next/Core and Builtins/2020-03-02-20-12-33.bpo-39776.fNaxi_.rst b/Misc/NEWS.d/next/Core and Builtins/2020-03-02-20-12-33.bpo-39776.fNaxi_.rst
new file mode 100644
index 000000000000..e5a00bd96ae4
--- /dev/null
+++ b/Misc/NEWS.d/next/Core and Builtins/2020-03-02-20-12-33.bpo-39776.fNaxi_.rst	
@@ -0,0 +1,6 @@
+Fix race condition where threads created by PyGILState_Ensure() could get a
+duplicate id.
+
+This affects consumers of tstate->id like the contextvar caching machinery,
+which could return invalid cached objects under heavy thread load (observed
+in embedded scenarios).
diff --git a/Python/pystate.c b/Python/pystate.c
index 90cd56bb05f4..ebfe6ce65de0 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -411,12 +411,12 @@ new_threadstate(PyInterpreterState *interp, int init)
         tstate->context = NULL;
         tstate->context_ver = 1;
 
-        tstate->id = ++interp->tstate_next_unique_id;
 
         if (init)
             _PyThreadState_Init(tstate);
 
         HEAD_LOCK();
+        tstate->id = ++interp->tstate_next_unique_id;
         tstate->prev = NULL;
         tstate->next = interp->tstate_head;
         if (tstate->next)