Hi,

bluhm@ has hit a bug earlier ([1]) that panics pretty late, after a
context switch to a different thread. The resulting backtrace is rather
useless. The problem was a missing mtx_leave() before calling
pool_get(..., PR_WAITOK) in a completely different code path.

This diff adds a check in pool_get() that we actually are in a sleepable
context when PR_WAITOK is given. There is an equivalent check present in
malloc() already.

Tests and feedback welcome.


 - Christian

[1] https://marc.info/?l=openbsd-bugs&m=175139531419854


diff --git a/sys/kern/subr_pool.c b/sys/kern/subr_pool.c
index d76e8b27cd57..7fb57d569d1c 100644
--- a/sys/kern/subr_pool.c
+++ b/sys/kern/subr_pool.c
@@ -584,8 +584,13 @@ pool_get(struct pool *pp, int flags)
 	}
 	pl_leave(pp, &pp->pr_lock);
 
-	if ((slowdown || pool_debug == 2) && ISSET(flags, PR_WAITOK))
-		yield();
+	if (ISSET(flags, PR_WAITOK)) {
+#ifdef DIAGNOSTIC
+		assertwaitok();
+#endif
+		if (slowdown || pool_debug == 2)
+			yield();
+	}
 
 	if (v == NULL) {
 		struct pool_get_memory mem = { .v = NULL };

• smime.p7s (5K)